fix: #2929 surface run-loop exceptions after stream_events() completes

[nileshpatil6]

Summary

Closes #2929.

When Runner.run_streamed() is used and the background run loop raises an exception early (e.g. during sandbox initialisation, before any stream event is produced), the failure could be silently swallowed — leaving callers with a result that looks clean (final_output=None, new_items=[]) but is actually broken.

This PR makes two minimal, additive changes to RunResultStreaming:

1. run_loop_exception property

A new read-only property that directly exposes the background task's exception after the stream ends, without requiring callers to reach into internal task state:

result = Runner.run_streamed(agent, "hello")
async for event in result.stream_events():
    pass

if result.run_loop_exception:
    raise result.run_loop_exception

Returns None if the run completed without error, has not yet finished, or was cancelled.

2. _check_errors() call in the stream_events() finally block

_await_task_safely() intentionally swallows all exceptions (they're supposed to be surfaced via _stored_exception). But _check_errors() was not called after _await_task_safely(run_loop_task) in the non-cancelled finally path. This meant a run-loop failure that raced past the sentinel was not transferred into _stored_exception, so the final if self._stored_exception: raise never fired.

Adding self._check_errors() after await self._await_task_safely(self.run_loop_task) closes this gap.

Changes

• src/agents/result.py — adds run_loop_exception property + _check_errors() call in stream_events() finally block
• tests/test_cancel_streaming.py — adds two tests:
  • test_run_loop_exception_property_is_none_on_success — property is None on a clean run
  • test_run_loop_exception_surfaced_after_stream — exception is both raised through stream_events() and accessible via the property

Test plan

• uv run pytest tests/test_cancel_streaming.py -v — 11 passed (9 existing + 2 new)
• uv run pyright src/agents/result.py — 0 errors