Don't let asyncio.run() shut down the DBOS executor - #925

Fixes #924

Description of the fix
Calling a DBOS async API from an event loop that DBOS doesn't own, like asyncio.run(), shuts down the DBOS executor when that loop closes. After that, every start_workflow and queue dispatch raised "cannot schedule new futures after shutdown" until the process restarted.

Implementation
_configure_asyncio_thread_pool() sets the DBOS executor itself as the running loop's default executor, and asyncio.run() calls shutdown() on that executor on exit. Now the loop gets a small ThreadPoolExecutor subclass that submits to the DBOS executor and ignores shutdown(). It has to be a subclass because set_default_executor() only accepts a ThreadPoolExecutor. It's created once per executor and reused, and DBOS.destroy() still shuts down the real executor like before.

Testing

• Added test_asyncio_run_keeps_dbos_executor to test_async.py. It calls a DBOS async API through asyncio.run(), then starts a sync workflow and runs an async workflow that uses asyncio.to_thread. On main, it fails with "cannot schedule new futures after shutdown". With this change, it passes on Postgres and SQLite.
• test_async, test_async_workflow_management, test_client, test_streaming, test_concurrency_primitives, test_dbos, test_queue and test_scheduler pass on Postgres 16 (417 passed, 1 skipped).
• black, isort, and mypy are clean, and pyright shows nothing new.