docs(mule-image): record uvicorn --reload removal + 2-worker switch

Backend was running --reload single-worker in production. Switched to
--workers 2 --proxy-headers. Also note the deploy.sh quirk where a
runtime-only command change needed --force-recreate to take effect.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

containers/120-mule-images.md

@@ -80,6 +80,13 @@ For pushes from inside the LXC, gitea creds at `/etc/mule-deploy/git-credentials
 
 ## Changelog
 
+### 2026-05-10 — backend out of dev mode
+`docker-compose.yml` backend `command` was running `uvicorn … --reload` (single worker, file-watcher). Page loads fanned out ~15 parallel API calls that all serialized on one event loop. Switched to `--workers 2 --proxy-headers`. Verified: `docker inspect mulita-backend --format '{{.Config.Cmd}}'` shows the new flags; two uvicorn worker processes visible.
+
+Gotcha: the `docker compose up -d --build` in `deploy.sh` did not recreate the backend after this commit — only `--build` rebuilds images, and the `command` change is runtime-only. Had to run `docker compose up -d --no-deps --force-recreate backend` manually. Consider adding `--force-recreate` to `deploy.sh` if compose-only changes become common.
+
+Still on the table for further `photos.hubris.network` perf work: pre-generate `/data/proxies` web-sized webps (currently empty volume → first open of any RAW/HEIC is a sync decode), shrink the 500-photo default page size, and back off `worker-status` / `scan/status` polling when idle.
+
 ### 2026-04-28 — wiki entry created
 Initial documentation.