feat: add embeddings pipeline and semantic search endpoint

Wire the full embedding flow:
- Rewrite Embedding model to use pgvector Vector(512) with HNSW index
- Add embed_photo, vision_fanout, backfill_vision Celery tasks on
  dedicated `vision` queue
- Hook vision_fanout into generate_thumbnails completion
- Add POST /api/v1/photos/search with hybrid RRF ranking (semantic-only
  for now; FTS leg added in PR5)
- Stub ocr_photo, detect_objects, extract_faces tasks for later PRs

Migration 0003 drops/recreates the embeddings table (was never populated).

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

backend/app/tasks/vision.py

@@ -0,0 +1,148 @@
+"""
+Celery tasks for the vision pipeline — embedding, OCR, object detection,
+face recognition.
+
+All tasks run on the dedicated `vision` queue with limited concurrency
+(memory-bound CPU inference). They read thumbnails generated by
+generate_thumbnails, so they MUST run after thumbs complete.
+"""
+import asyncio
+import logging
+from pathlib import Path
+
+import numpy as np
+from celery import shared_task
+from sqlalchemy import select, delete, text
+from PIL import Image
+
+from app.database import AsyncSessionLocal
+from app.models import Photo
+from app.models.embeddings import Embedding
+from app.config import settings
+
+logger = logging.getLogger(__name__)
+
+
+def _load_thumb(photo_id: str, size: str = "medium") -> np.ndarray | None:
+    """Load a thumbnail as an RGB numpy array."""
+    thumb_path = Path(f"/data/thumbs/{photo_id}/{size}.webp")
+    if not thumb_path.exists():
+        logger.warning("Thumbnail not found: %s", thumb_path)
+        return None
+    img = Image.open(thumb_path).convert("RGB")
+    return np.array(img)
+
+
+@shared_task(name='embed_photo', queue='vision')
+def embed_photo(photo_id: str):
+    """Generate CLIP embedding for a photo and store in pgvector."""
+    if not settings.vision.enabled:
+        return {'status': 'skipped', 'reason': 'vision disabled'}
+    return asyncio.run(_embed_photo_async(photo_id))
+
+
+async def _embed_photo_async(photo_id: str):
+    image = _load_thumb(photo_id, "medium")  # 640px
+    if image is None:
+        return {'status': 'error', 'message': 'thumbnail not found'}
+
+    from app.services.vision.registry import registry
+    embedder = registry.get_embedder()
+    vector = embedder.embed_image(image)
+
+    model_name = settings.vision.embedder.name
+
+    async with AsyncSessionLocal() as session:
+        # Upsert: delete existing then insert
+        await session.execute(
+            delete(Embedding).where(
+                Embedding.photo_id == photo_id,
+                Embedding.model == model_name,
+            )
+        )
+        emb = Embedding(
+            photo_id=photo_id,
+            model=model_name,
+            vector=vector.tolist(),
+        )
+        session.add(emb)
+        await session.commit()
+
+    logger.info("Embedded photo %s with %s", photo_id, model_name)
+    return {'status': 'success', 'photo_id': photo_id}
+
+
+@shared_task(name='vision_fanout', queue='vision')
+def vision_fanout(photo_id: str):
+    """Dispatch all enabled vision tasks for a photo."""
+    if not settings.vision.enabled:
+        return {'status': 'skipped', 'reason': 'vision disabled'}
+
+    embed_photo.delay(photo_id)
+
+    if settings.vision.ocr.enabled:
+        ocr_photo.delay(photo_id)
+    if settings.vision.detector.enabled:
+        detect_objects.delay(photo_id)
+    if settings.vision.faces.enabled:
+        extract_faces.delay(photo_id)
+
+    return {'status': 'dispatched', 'photo_id': photo_id}
+
+
+@shared_task(name='ocr_photo', queue='vision')
+def ocr_photo(photo_id: str):
+    """OCR a photo — implemented in PR5."""
+    return {'status': 'not_implemented'}
+
+
+@shared_task(name='detect_objects', queue='vision')
+def detect_objects(photo_id: str):
+    """Detect objects in a photo — implemented in PR6."""
+    return {'status': 'not_implemented'}
+
+
+@shared_task(name='extract_faces', queue='vision')
+def extract_faces(photo_id: str):
+    """Detect faces and extract embeddings — implemented in PR7."""
+    return {'status': 'not_implemented'}
+
+
+@shared_task(name='backfill_vision')
+def backfill_vision(task: str | None = None, limit: int | None = None):
+    """Queue vision tasks for photos that haven't been processed yet."""
+    return asyncio.run(_backfill_vision_async(task, limit))
+
+
+async def _backfill_vision_async(task: str | None, limit: int | None):
+    model_name = settings.vision.embedder.name
+
+    async with AsyncSessionLocal() as session:
+        # Find photos without embeddings
+        stmt = text("""
+            SELECT p.id FROM photos p
+            LEFT JOIN embeddings e ON e.photo_id = p.id AND e.model = :model
+            WHERE e.photo_id IS NULL
+            AND p.processing_status = 'completed'
+            ORDER BY p.created_at DESC
+        """)
+        if limit:
+            stmt = text(str(stmt) + f" LIMIT {limit}")
+
+        result = await session.execute(stmt, {"model": model_name})
+        photo_ids = [row[0] for row in result.fetchall()]
+
+    count = 0
+    for pid in photo_ids:
+        if task == 'embed' or task is None:
+            embed_photo.delay(pid)
+        if task == 'ocr' or task is None:
+            ocr_photo.delay(pid)
+        if task == 'detect' or task is None:
+            detect_objects.delay(pid)
+        if task == 'faces' or task is None:
+            extract_faces.delay(pid)
+        count += 1
+
+    logger.info("Backfill queued %d photos for vision processing", count)
+    return {'status': 'queued', 'count': count}