fix: image search 500 on Qdrant Cloud

- Missing collection_id payload index on the image collection — Qdrant
  Cloud rejects filters on unindexed fields (local docker tolerated it)
- ensure_collections now runs at app startup, not just first ingest

Author: MeKaustubh07

backend/app/main.py

@@ -1,3 +1,4 @@
+import asyncio
 from contextlib import asynccontextmanager
 
 from arq import create_pool
@@ -7,12 +8,15 @@
 
 from app.api import chat, collections, documents, health
 from app.core.config import get_settings
+from app.retrieval.vectorstore import ensure_collections
 
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
     settings = get_settings()
     app.state.arq = await create_pool(RedisSettings.from_dsn(settings.redis_url))
+    # collections + payload indexes must exist before any search hits them
+    await asyncio.to_thread(ensure_collections)
     yield
     await app.state.arq.close()
 

backend/app/retrieval/vectorstore.py

@@ -49,6 +49,9 @@ def ensure_collections() -> None:
         client.create_payload_index(
             IMAGE_COLLECTION, "tenant_id", models.PayloadSchemaType.KEYWORD
         )
+        client.create_payload_index(
+            IMAGE_COLLECTION, "collection_id", models.PayloadSchemaType.KEYWORD
+        )
 
 
 def _tenant_filter(tenant_id: str, collection_id: str | None) -> models.Filter: