Run each async udf in its own thread

[nkitsaini]

Note

Medium Risk
Touches core external-function request scheduling, concurrency, and cancellation paths; misbehavior could affect timeouts, client disconnects, or async UDF stability under load.

Overview
Async UDF execution no longer runs inline on the ASGI request loop (or via a per-call asyncio.run in a thread for async paths). Async invocations are scheduled on a process-wide dedicated event-loop thread via _dispatch_to_async_loop and _cancellable_run, so loop-bound clients (e.g. httpx) can be reused and concurrent async calls are not serialized. Sync UDFs are explicitly routed through to_thread + asyncio.run so they cannot block that shared loop.

Cancellation is reworked: cancel_event is set when the function task is still pending after the first-completed wait, and always in finally, instead of only on disconnect/timeout paths—so off-thread work actually sees the signal.

Managed UDF server (run_udf_app) sets Uvicorn timeout_keep_alive from SINGLESTOREDB_UDF_KEEPALIVE_TIMEOUT (default 120 seconds) to reduce aggressive idle connection closes under load.

Adds test_udf_event_loop.py covering cancellable runs, dispatch concurrency, resource reuse, and Application routing (async on dispatch thread vs sync on a worker).

^{Reviewed by Cursor Bugbot for commit 0a7a0a9. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 018db7a. Configure here.}