Allow splitting config, downloads and media paths via .env

Adds optional CONFIG_DIR, DOWNLOADS_DIR and MEDIA_DIR env variables so
users can spread configs, downloads and media across different drives
without restructuring ROOT_DIR. Each falls back to its ROOT_DIR-based
default when left empty, so existing setups keep working unchanged.

The -arr services (sonarr, radarr, lidarr, mylar3) now use a single
${DOWNLOADS_DIR}:/data parent bind with a ${MEDIA_DIR}:/data/media
overlay. This preserves TRaSH's single-/data-mount semantics inside
the container (hardlinks still work on one filesystem) while letting
the host-side media tree live wherever the user wants.

README documents the new variables, the generated folder structure,
recommended storage layouts, and the same-filesystem requirement for
hardlinks between downloads and media.

README.md

@@ -55,9 +55,13 @@ For other Linux distros you may have to use a different package manager or downl
 ### Using the CLI
 To make things easier, a CLI has been developed. First, clone the repository in a directory of your
 choosing. You can run it by entering `python3 main.py` and the CLI will guide you through the
-process. This is the recommended method if you're setting this up for the first time on a new system. 
-Please take a look at [important notes](#important-notes) before you continue. 
-**NOTE: This script will create users for each container with IDs ranging from 13001 to 13014. 
+process. It will ask which services to enable, your timezone, your root directory, and — if you
+want — independent absolute paths for your configs, downloads and media trees (press enter at each
+prompt to keep the `ROOT_DIR`-based default). See [Storage layouts](#storage-layouts) below for the
+trade-offs when splitting those.
+This is the recommended method if you're setting this up for the first time on a new system.
+Please take a look at [important notes](#important-notes) before you continue.
+**NOTE: This script will create users for each container with IDs ranging from 13001 to 13014.
 If you want to choose your own IDs (or some of them are occupied) you have to go through the manual install.**
 
 ### Manually
@@ -65,9 +69,55 @@ If you're installing this for the first time simply follow these steps.
 If you're coming from an older version or reinstalling with different IDs, run `remove_old_users.sh` to clean up old users and then follow these steps.
 1. To get started, clone the repository in a directory of your choosing. `git clone https://github.com/Luctia/ezarr.git`
 2. Copy `.env.sample` to a real `.env` by running `$ cp .env.sample .env`.
-3. Set the environment variables to your liking. Pay special attention `ROOT_DIR` as this is where everything is going to be stored in. 
+3. Set the environment variables to your liking. Pay special attention `ROOT_DIR` as this is where everything is going to be stored in.
    The path in this value needs to be **absolute**. If you leave it empty it's going to install in the directory the .env file is currently in.
    `UID` should be set to the ID of the user that you want to run docker with. You can find this by running `id -u` from that user's shell.
+   If you want to spread configs, downloads and media across different locations (e.g. SSD for configs, bulk HDD for media), set `CONFIG_DIR`, `DOWNLOADS_DIR` and/or `MEDIA_DIR` to absolute paths. Any of them left empty fall back to `${ROOT_DIR}/config`, `${ROOT_DIR}/data` and `${ROOT_DIR}/data/media` respectively. **Note:** hardlinks/atomic moves between downloads and media only work when `DOWNLOADS_DIR` and `MEDIA_DIR` live on the same filesystem — split them only if you understand that you'll be copying instead of hardlinking.
+
+   The generated folder structure on the host looks like this (the three top-level trees can live under one `ROOT_DIR` or be spread across different drives via the `*_DIR` overrides):
+   ```
+   ${CONFIG_DIR}/                # default: ${ROOT_DIR}/config
+     ├── sonarr-config/
+     ├── radarr-config/
+     ├── lidarr-config/
+     ├── mylar-config/
+     ├── prowlarr-config/
+     ├── jackett-config/
+     ├── qbittorrent-config/
+     ├── sabnzbd-config/
+     ├── bazarr-config/
+     ├── overseerr-config/
+     ├── jellyseerr-config/
+     ├── plex-config/
+     ├── jellyfin-config/
+     ├── tautulli-config/
+     └── audiobookshelf-config/
+
+   ${DOWNLOADS_DIR}/             # default: ${ROOT_DIR}/data
+     ├── torrents/               # qBittorrent save path
+     │   ├── tv/
+     │   ├── movies/
+     │   ├── music/
+     │   ├── comics/
+     │   └── ...
+     └── usenet/                 # SABnzbd complete/incomplete
+         ├── tv/
+         ├── movies/
+         ├── music/
+         ├── comics/
+         └── ...
+
+   ${MEDIA_DIR}/                 # default: ${ROOT_DIR}/data/media
+     ├── tv/                     # Sonarr
+     ├── movies/                 # Radarr
+     ├── music/                  # Lidarr
+     ├── comics/                 # Mylar3
+     ├── books/
+     ├── audiobooks/             # Audiobookshelf
+     ├── podcasts/               # Audiobookshelf
+     └── audiobookshelf-metadata/
+   ```
+   Inside the -arr containers these are always exposed under a single `/data` tree — `/data/torrents`, `/data/usenet`, `/data/media/{tv,movies,...}` — regardless of how the host paths are split. This preserves the TRaSH single-`/data`-mount pattern and keeps hardlinks working as long as `DOWNLOADS_DIR` and `MEDIA_DIR` share a filesystem. The compose file achieves this by binding `${DOWNLOADS_DIR}:/data` as the parent and overlaying `${MEDIA_DIR}:/data/media` on top.
 4. Run `setup.sh` as superuser. This will set up your users, a system of directories and ensure permissions are set correctly.
 5. Copy `docker-compose.yml.sample` to a real `docker-compose.yml` by running `$ cp docker-compose.yml.sample docker-compose.yml`.
 6. Take a look at the `docker-compose.yml` file. If there are services you would like to ignore
@@ -86,6 +136,41 @@ If you're coming from an older version or reinstalling with different IDs, run `
 That's it! Your containers are now up and you can continue to set up the settings in them. Please
 take a look at [important notes](#important-notes) before you continue.
 
+## Storage layouts
+The -arr services use **hardlinks** to move downloaded files from `torrents/` or `usenet/` into the
+media library without duplicating bytes. Hardlinks only work within a single filesystem — so
+`DOWNLOADS_DIR` and `MEDIA_DIR` must resolve to paths on the same filesystem for this to work. If
+they don't, Sonarr/Radarr/Lidarr/Mylar will silently fall back to copying, which uses twice the
+disk space and a lot more I/O on every import. `CONFIG_DIR` has no such constraint and can live
+anywhere.
+
+Some layouts that work well:
+
+- **Single-drive / single-filesystem (simplest).** Set `ROOT_DIR` and leave `CONFIG_DIR`,
+  `DOWNLOADS_DIR`, `MEDIA_DIR` empty. Everything lives under one tree, hardlinks just work.
+- **SSD for configs, HDD for everything else.** Set `CONFIG_DIR=/ssd/ezarr/config` and leave the
+  other two empty (they default under `ROOT_DIR`, which you point at the HDD). Config databases
+  benefit from the faster disk; downloads + media stay together so hardlinks work.
+- **Pooled storage with [mergerfs](https://github.com/trapexit/mergerfs).** Combine multiple
+  HDDs into one mount point (e.g. `/pool`) and point `ROOT_DIR` or `DOWNLOADS_DIR`+`MEDIA_DIR` at
+  it. With `category.create=epmfs` and `moveonenospc=true`, mergerfs keeps a torrent and its
+  hardlinked media file on the same underlying disk, so `link(2)` succeeds inside the pool.
+- **CoW filesystems (btrfs / xfs with reflink / bcachefs / ZFS).** Same-filesystem rule still
+  applies, but you also get reflinks (instant, space-sharing copies) as a fallback when hardlinks
+  aren't appropriate.
+
+Layouts that will **break hardlinks** (they will still run, just less efficiently):
+- `DOWNLOADS_DIR` on one drive/mount and `MEDIA_DIR` on a different drive/mount.
+- Downloads on local disk, media on a remote NFS/SMB share (or vice versa).
+- Either path going through a union filesystem that doesn't co-locate related files.
+
+If you genuinely need split storage and can't pool it, set "Use Hardlinks" to `false` in each -arr
+and accept the copy overhead.
+
+Regardless of how you split the host paths, the compose file always exposes a single `/data` tree
+inside the -arr containers — `/data/torrents`, `/data/usenet`, `/data/media/{tv,movies,...}` — so
+the root-folder paths you configure in Sonarr/Radarr/etc. never need to change.
+
 ## Important notes
 - You probably shouldn't run the python script as root. Ideally you should create a brand new user that's just for these services, but any regular user will do.
   It will need your password for `sudo` to set up the permissions and folder structures, but you shouldn't run it *as* root.
@@ -114,11 +199,13 @@ Note that if you are a group member (and not the owner), having matching group I
 the users and groups are set up on both sides correctly is to run `setup.sh` on both your NFS server and your client. 
 On your server:
 - Copy `.env` and `setup.sh` to your NFS server.
-- You may have to adjust `.env` so that `ROOT_DIR` reflects where it will be stored on your server, which is most likely different from the mapped location on the client.
+- You may have to adjust `.env` so that `ROOT_DIR` (and `CONFIG_DIR` / `DOWNLOADS_DIR` / `MEDIA_DIR`, if you've set any of them) reflect where each tree will be stored on your server, which is most likely different from the mapped location on the client.
 - Make sure that the `.env` file is not a .sample. Run `setup.sh`.
-- Now follow all the same steps but on your client machine. Always double-check that `.env` is set correctly, especially `ROOT_DIR`.
-You don't have to do this on your server first but it's recommended. If you are running this script on the client **make sure that you temporarily enable -no-root-squash on your NFS server**, 
+- Now follow all the same steps but on your client machine. Always double-check that `.env` is set correctly, especially the path variables above.
+You don't have to do this on your server first but it's recommended. If you are running this script on the client **make sure that you temporarily enable -no-root-squash on your NFS server**,
 as the script needs superuser privileges to run and by default on NFS the root user is mapped to nowhere to prevent abuse.
+
+**Heads-up on hardlinks over NFS:** if your downloads live on the local disk but media lives on an NFS mount (or vice versa), that's two different filesystems and hardlinks will break — see [Storage layouts](#storage-layouts) for the options. Keeping `DOWNLOADS_DIR` and `MEDIA_DIR` both on the NFS mount is fine, as long as they share the same share/filesystem.
   
 ### SABnzbd External internet access denied message
 When you're trying to access SABnzbd the first time you'll come across the message `External