fix(asyncio): preserve producer failures while draining consumers

[sylvesterkaczmarek]

Summary

• preserve the producer exception when the consumer also fails during producer-failure draining
• continue waiting for the consumer task to settle before propagating the producer failure
• keep consumer-first failure and parent-cancellation behaviour unchanged

run_producer_consumer() documents producer failures as primary while the consumer drains. Previously a secondary consumer exception raised during that drain replaced the producer exception.

Test plan

• added focused regression coverage in tests/test_asyncio_tasks_primary_error.py
• GitHub Actions

Issue number

N/A

[seratch]

I traced run_producer_consumer() through its two current callers. Both dispatchers already catch and log ordinary Exceptions from on_stream, so the test's ConsumerError(Exception) cannot reach this helper through the supported Agent.as_tool() or Codex tool paths. The consumer failures that intentionally escape are BaseExceptions such as CancelledError; preserving that behavior was part of the #4198 and #4201 fix.

In a controlled Agent.as_tool(on_stream=...) interleaving, the released code propagates CancelledError, while this patch replaces it with the producer RuntimeError. With the default failure handler, that becomes a normal tool-error result and can let the outer run continue. Since no supported scenario demonstrates why producer failure should override consumer cancellation, I do not think this change is worth merging. I am going to close this PR for now.