Summary

Fixes three related bugs in networkloop.py caused by stop_event being partially ignored during shutdown:

• Shutdown hangs: network_retry_loop used a bare asyncio.sleep during its retry backoff, ignoring stop_event entirely. This caused shutdown routines to hang for up to 30 seconds during network outages. Replaced with asyncio.wait_for(stop_event.wait(), timeout=...).
• Orphaned tasks: Pending tasks (such as in-flight HTTP calls) were cancelled but never awaited before do_action returned, leading to "Task was destroyed but it is pending!" garbage collector warnings. Added asyncio.gather(*pending, return_exceptions=True) to ensure clean teardown.
• Runaway loop: When repeat_on_success=True, the loop failed to break if stop_event was set during an action. The loop would immediately restart and spin indefinitely. Added explicit stop_event.is_set() exit checks to break the outer loop safely.

Adds three regression tests to prevent recurrence.

Tests

uv run --group tests pytest tests/ext/_utils/test_networkloop.py -v

Results:

tests/ext/_utils/test_networkloop.py::TestNetworkRetryLoop::test_stop_event_interrupts_backoff_sleep PASSED [ 87%]
tests/ext/_utils/test_networkloop.py::TestNetworkRetryLoop::test_stop_event_breaks_repeat_on_success_loop PASSED [ 93%]
tests/ext/_utils/test_networkloop.py::TestNetworkRetryLoop::test_no_pending_tasks_after_stop_event PASSED [100%]

============================= 16 passed in 13.90s ==============================

Live verification:

Ran a custom asyncio verification script locally simulating network outages. Confirmed the event loop exits promptly on shutdown and no pending task warnings are emitted during teardown.