Fix event loop closed warning during stdio cleanup

[jlowin]

When using Client with stdio transport in pytest, an "Event loop is closed" warning can appear during test teardown on Linux CI. This happens because asyncio's BaseSubprocessTransport.__del__ gets triggered by GC after pytest-asyncio closes the event loop.

This PR attempts to fix the issue by forcing GC and yielding to the event loop after the stdio_client context manager exits but while the loop is still running. This should allow any subprocess transport cleanup callbacks to execute before the loop closes.

# After stdio_client context exits, force garbage collection while the
# event loop is still running.
gc.collect()
await asyncio.sleep(0)

Note: This fix may not fully resolve the issue - it needs to be tested in CI on Linux. The warning is fundamentally a race between GC and event loop closure, so there may still be edge cases. If this doesn't work, we may need to either:

1. Make the fix more aggressive (multiple GC cycles, longer delay)
2. Suppress the warning in pytest configuration
3. Report upstream to the MCP SDK or Python

Closes #2792

---

Tests added:

• tests/client/test_stdio_cleanup.py - Tests that convert PytestUnraisableExceptionWarning to errors to catch the issue