fix: #3334 await realtime background tasks during cleanup

[Aphroq]

Summary

• Await cancelled Realtime guardrail and tool-call background tasks during session cleanup before clearing their tracking sets.
• Keep cleanup exception handling exercised with real asyncio tasks, and add a regression test that verifies cancelled tasks finish their finally blocks before _cleanup() returns.

Related: #1976 previously identified the missing await for Realtime guardrail task cleanup. This PR keeps that direction, also covers Realtime tool-call task cleanup, and adds regression coverage.

Test plan

• uv run pytest tests/realtime/test_session.py::test_cleanup_awaits_cancelled_background_tasks tests/realtime/test_session_exceptions.py::TestSessionExceptions::test_exception_during_guardrail_processing
• bash .agents/skills/code-change-verification/scripts/run.sh

Issue number

Closes #3334

Checks

• I've added new tests (if relevant)
• I've added/updated the relevant documentation
• I've run make lint and make format
• I've made sure tests pass

[Aphroq]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4f50cdfb6b

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Bound cleanup waits for cancelled user tasks

When a realtime tool call is running user code that suppresses or delays CancelledError, close()/__aexit__ now waits indefinitely here before closing the model connection or waking event iterators. Tool functions are arbitrary async callables, so a cancellation-resistant tool can make an async with RealtimeSession(...) hang during shutdown; the guardrail cleanup uses the same unbounded gather pattern. Consider bounding the wait or otherwise ensuring cleanup can finish even when a background task does not promptly exit after cancellation.

Useful? React with 👍 / 👎.