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/ocr.py

@@ -0,0 +1,45 @@
+"""
+OCR engine using rapidocr-onnxruntime (PP-OCRv4 weights).
+
+No PaddlePaddle dependency — pure ONNX Runtime. Language packs are
+downloaded automatically by rapidocr on first use.
+"""
+import logging
+
+import numpy as np
+
+from app.config import VisionSettings
+from app.services.vision.base import OCREngine, OCRResult
+
+logger = logging.getLogger(__name__)
+
+
+class RapidOCREngine(OCREngine):
+    def __init__(self, settings: VisionSettings):
+        from rapidocr_onnxruntime import RapidOCR
+
+        self._min_confidence = settings.ocr.min_confidence
+        self._engine = RapidOCR()
+        logger.info("RapidOCR engine initialized")
+
+    def run(self, image: np.ndarray) -> list[OCRResult]:
+        result, _ = self._engine(image)
+        if not result:
+            return []
+
+        out = []
+        for box, text, score in result:
+            if score < self._min_confidence:
+                continue
+            # box is [[x1,y1],[x2,y2],[x3,y3],[x4,y4]] — take bounding rect
+            xs = [p[0] for p in box]
+            ys = [p[1] for p in box]
+            h, w = image.shape[:2]
+            bbox = [
+                min(xs) / w,
+                min(ys) / h,
+                max(xs) / w,
+                max(ys) / h,
+            ]
+            out.append(OCRResult(text=text, confidence=float(score), bbox=bbox))
+        return out