Async AI replies for cloud scale
Long-running AI replies now run on a background worker queue instead of holding a web worker, so the chat widget stays snappy under load.
Includes a synchronous fallback and a per-workspace toggle, with the exact prompt and response still recorded in AI History.