Separate ASR (2 workers) from summary/RAG post-processing, add LightRAG chat API, batch upload fixes, and local model mounts for Docker deployment. Co-authored-by: Cursor <cursoragent@cursor.com>
