perf(warm): warm the full search path at startup, not just the model

coderag/api.py

@@ -196,13 +196,23 @@ def delete_path(self, path: Union[str, Path]) -> int:
             return self.store.delete_file(rel)
 
     def warm(self) -> None:
-        """Eagerly load the provider, store, and embedding model.
+        """Eagerly load the provider, store, and embedding model — and the search path.
 
         Done at server startup so the first query — and the demo UI's search-speed
-        badge — reflect warm performance, not the one-off lazy model load.
+        badge — reflect warm performance, not the one-off lazy load. A real search is
+        run (not just an embed) because the store's vector/FTS/scalar indexes and
+        LanceDB's query path are loaded lazily on first use; warming only the model
+        leaves that cold-load to land on the first user query, where it shows up as a
+        large ``store_ms``. Best-effort: warm-up failures must not block startup.
         """
         self.status()  # builds provider/store
         self.provider.embed_query("warm up")  # loads the model + JITs the query path
+        try:
+            # Exercise the full retrieval path (vector + lexical + hydrate) so the
+            # store's indexes are resident before the first real query.
+            self.search("warm up", top_k=1)
+        except Exception:  # pragma: no cover - warm-up is best-effort
+            logger.exception("Search warm-up failed (continuing).")
 
     def status(self) -> dict:
         """Index statistics and provenance."""