fix(semantic): batch VLM summary generation to prevent event-loop deadlock

openviking/storage/queuefs/semantic_processor.py

@@ -491,7 +491,18 @@ async def _gen(idx: int, file_path: str) -> None:
                     logger.warning(f"Failed to generate summary for {file_path}: {e}")
                     file_summaries[idx] = {"name": file_name, "summary": ""}
 
-            await asyncio.gather(*[_gen(i, fp) for i, fp in pending_indices])
+            # Process in batches to prevent event-loop starvation when the
+            # pending list is large (e.g. hundreds of files).  Without batching
+            # all coroutines are spawned at once via asyncio.gather, which can
+            # deadlock the event loop even with a semaphore in place.
+            _BATCH_SIZE = 20
+            for _batch_start in range(0, len(pending_indices), _BATCH_SIZE):
+                _batch = pending_indices[_batch_start : _batch_start + _BATCH_SIZE]
+                await asyncio.gather(*[_gen(i, fp) for i, fp in _batch])
+                logger.info(
+                    f"Summary batch {_batch_start // _BATCH_SIZE + 1}/"
+                    f"{(len(pending_indices) + _BATCH_SIZE - 1) // _BATCH_SIZE} done"
+                )
 
         file_summaries = [s for s in file_summaries if s is not None]