perf: enable multi-threaded video decoding and offload file persistence to background thread

app.py

@@ -160,6 +160,7 @@ def load_video(video_path, max_frames=64):
     try:
         container = av.open(video_path)
         stream = container.streams.video[0]
+        stream.thread_count = 8 # Enable multi-threaded decoding
         duration = stream.duration
         if duration is None or duration <= 0:
             # Fallback to full decode for short/broken streams
@@ -248,9 +249,9 @@ def predict(
     Streaming inference endpoint with history support.
     """
     session_id = str(uuid.uuid4())
-    # Persist files first for audit parity
+    # Persist files in background to avoid blocking user (parity audit)
     if files:
-        persist_uploaded_files(files, session_id)
+        threading.Thread(target=persist_uploaded_files, args=(files, session_id), daemon=True).start()
     messages = []
     
     # Process history