feat: add vision pipeline scaffolding with ONNX backend

Introduce the app/services/vision/ module with ABC interfaces, ONNX
Runtime backend, model registry, and per-task implementations:
- OpenCLIP ViT-B/32 embedder (image + text, 512-d)
- RapidOCR engine (PP-OCRv4 via ONNX, no PaddlePaddle)
- YOLOv8n object detector (raw ONNX, no ultralytics runtime)
- YuNet + SFace face processor (Apache 2.0, opencv_zoo, 128-d)
- DBSCAN face clustering helper

Add VisionSettings to config (mulita.yml + Pydantic), bootstrap_models.py
for first-boot weight downloads, models_data Docker volume, and ROCm
backend stub for future GPU acceleration.

No Celery tasks wired yet — models load but nothing invokes them.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

backend/app/services/vision/bootstrap_models.py

@@ -0,0 +1,83 @@
+"""
+Download vision model weights on first worker boot.
+
+Run as:  python -m app.services.vision.bootstrap_models
+
+Or called from the vision worker entrypoint before Celery starts.
+Downloads are idempotent — existing files with matching sizes are skipped.
+"""
+import logging
+import os
+from pathlib import Path
+from urllib.request import urlretrieve
+
+from app.config import settings
+
+logger = logging.getLogger(__name__)
+
+# (relative_path, url, expected_size_bytes_approx)
+# Sizes are approximate — used only for skip-if-exists checks, not integrity.
+MODELS = [
+    # OpenCLIP ViT-B/32 — visual and textual encoders (ONNX)
+    # These must be exported manually via export_openclip.py (see below).
+    # Placeholder entries — bootstrap will warn if missing.
+    ("embed/visual.onnx", None, None),
+    ("embed/textual.onnx", None, None),
+
+    # YOLOv8n — object detection
+    # Export: `yolo export model=yolov8n.pt format=onnx imgsz=640`
+    # Placeholder — must be exported from ultralytics offline.
+    ("detect/yolov8n.onnx", None, None),
+
+    # YuNet — face detection (Apache 2.0, opencv_zoo)
+    (
+        "face/yunet.onnx",
+        "https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/face_detection_yunet_2023mar.onnx",
+        233_000,
+    ),
+
+    # SFace — face recognition (Apache 2.0, opencv_zoo)
+    (
+        "face/sface.onnx",
+        "https://github.com/opencv/opencv_zoo/raw/main/models/face_recognition_sface/face_recognition_sface_2021dec.onnx",
+        37_000_000,
+    ),
+]
+
+
+def bootstrap(models_dir: str | None = None):
+    """Ensure all model files are present. Download what we can, warn about
+    files that need manual export."""
+    base = Path(models_dir or settings.vision.models_dir)
+    base.mkdir(parents=True, exist_ok=True)
+
+    for rel_path, url, expected_size in MODELS:
+        dest = base / rel_path
+        dest.parent.mkdir(parents=True, exist_ok=True)
+
+        if dest.exists():
+            logger.debug("Model already exists: %s", dest)
+            continue
+
+        if url is None:
+            logger.warning(
+                "Model file %s not found and has no auto-download URL. "
+                "See bootstrap_models.py for export instructions.",
+                dest,
+            )
+            continue
+
+        logger.info("Downloading %s → %s", url, dest)
+        try:
+            urlretrieve(url, str(dest))
+            actual = dest.stat().st_size
+            logger.info("Downloaded %s (%d bytes)", rel_path, actual)
+        except Exception as e:
+            logger.error("Failed to download %s: %s", rel_path, e)
+            if dest.exists():
+                dest.unlink()
+
+
+if __name__ == "__main__":
+    logging.basicConfig(level=logging.INFO)
+    bootstrap()