fix: revert conditional skip that caused no_docs regression

The conditional skip optimization broke vector search because query_doc_with_hybrid_search requires collection_result for validation even when BM25 is disabled.\n\nChanges:\n- Always fetch collection data (required for validation)\n- Keep sync fetch optimization for single collections\n- Simplify task loop filter

backend/open_webui/retrieval/utils.py

@@ -544,28 +544,20 @@ async def query_collection_with_hybrid_search(
     results = []
     error = False
 
-    # Optimization: Only fetch collection data if explicitly needed for BM25 or enrichment.
-    # Fetching massive collections (e.g. 160k docs) takes significant time (~30s) and memory.
-    should_fetch_data = hybrid_bm25_weight > 0 or enable_enriched_texts
-
-    if should_fetch_data:
-        # Optimization: If only one collection, fetch synchronously to avoid asyncio.to_thread overhead
-        # which can cause significant regression (e.g. 4s vs 28s) for large payloads.
-        if len(collection_names) == 1:
-            try:
-                col_name = collection_names[0]
-                log.debug(f"Fetching single collection synchronously: {col_name}")
-                data = VECTOR_DB_CLIENT.get(collection_name=col_name)
-                collection_results = {col_name: data}
-            except Exception as e:
-                log.exception(f"Failed to fetch collection {col_name}: {e}")
-                collection_results = {col_name: None}
-        else:
-            # Fetch all collections in parallel to reduce latency for multiple collections
-            collection_results = await fetch_collection_data(collection_names)
+    # Always fetch collection data - it's required for validation in query_doc_with_hybrid_search.
+    # Optimization: Use sync fetch for single collection to avoid asyncio overhead.
+    if len(collection_names) == 1:
+        try:
+            col_name = collection_names[0]
+            log.debug(f"Fetching single collection synchronously: {col_name}")
+            data = VECTOR_DB_CLIENT.get(collection_name=col_name)
+            collection_results = {col_name: data}
+        except Exception as e:
+            log.exception(f"Failed to fetch collection {col_name}: {e}")
+            collection_results = {col_name: None}
     else:
-        log.debug("Skipping collection fetch: BM25 disabled and no text enrichment needed")
-        collection_results = {name: None for name in collection_names}
+        # Fetch all collections in parallel to reduce latency for multiple collections
+        collection_results = await fetch_collection_data(collection_names)
 
     log.info(
         f"Starting hybrid search for {len(queries)} queries in {len(collection_names)} collections..."
@@ -591,16 +583,13 @@ async def query_collection_with_hybrid_search(
             return None, e
 
     # Prepare tasks for all collections and queries
-    # If we skipped fetching (should_fetch_data=False), we still want to run the query (using vector search only).
-    # If we tried to fetch and failed (result is None), we skip.
-    tasks = []
-    for collection_name in collection_names:
-        data = collection_results.get(collection_name)
-        if data is None and should_fetch_data:
-             # We needed data but failed to fetch it, skip this collection
-             continue
-        for query in queries:
-             tasks.append((collection_name, query))
+    # Skip collections that failed to fetch (data is None)
+    tasks = [
+        (collection_name, query)
+        for collection_name in collection_names
+        if collection_results.get(collection_name) is not None
+        for query in queries
+    ]
 
     # Run all queries in parallel using asyncio.gather
     task_results = await asyncio.gather(