fix(core): prevent listener crashes on late CDP transaction responses

[Avejack]

Summary

This PR fixes a race condition where late CDP responses for cancelled/completed transactions could raise InvalidStateError and terminate Listener.listener_loop(), causing automation sessions to become unhealthy over long runtimes.

Problem

In Transaction.__call__, response completion used set_result() / set_exception() directly without guarding future state.
When a response arrived after cancellation/completion, asyncio.InvalidStateError could be raised.
That exception propagated from tx(**message) in listener_loop, killing the listener task.

Root Cause

• Transaction.__call__ had no done-state guard.
• No safe handling for InvalidStateError during future completion.
• listener_loop invoked transaction completion without isolation.
• Connection.send() and _send_oneshot() could leave stale mapper entries on cancellation.

Changes

• Updated zendriver/core/connection.py.
• Added done-state guard in Transaction.__call__.
• Added safe completion helpers:
  • _set_result_safely
  • _set_exception_safely
• Converted parse-failure path to set transaction exception instead of raising out.
• Added missing-response guard ("result" not present).
• Added cancellation cleanup in Connection.send() mapper lifecycle.
• Added cancellation/finally cleanup in Connection._send_oneshot().
• Added Listener._complete_transaction() wrapper to prevent listener crash on completion errors.
• Updated listener response paths to use _complete_transaction.
• Added changelog entry in CHANGELOG.md under Unreleased -> Fixed.
• Added regression tests in tests/core/test_connection.py:
  • test_transaction_late_response_after_cancel_does_not_raise
  • test_transaction_parse_mismatch_sets_exception_not_raise
  • test_send_cancellation_removes_tx_from_mapper
  • test_listener_complete_transaction_handles_exceptions

Behavior Impact

• No public API changes.
• Internal reliability improvement only.
• Late/stale responses are safely ignored or converted into transaction-level errors.
• Listener task remains alive when transaction completion encounters unexpected errors.

Validation

Executed:

• poetry run pytest -q tests/core/test_connection.py

Result:

• 4 passed

Notes

A pre-existing warning remains unrelated to this fix:

• zendriver/core/connection.py warns about continue in a finally block.

Risk Assessment

Low.
Changes are scoped to internal transaction/listener robustness and covered by new regression tests.

Pre-merge Checklist

• I have described my change in the section above.
• I have ran the ./scripts/format.sh and ./scripts/lint.sh scripts. My code is properly formatted and has no linting errors.
• I have ran uv run pytest and ensured all tests pass.
• I have added my change to CHANGELOG.md under the [Unreleased] section.

[stephanlensky]

Thanks for the contribution @Avejack, the changes make sense to me.

Looks like the lint is failing, would you mind taking a look? I'd suggest using the included scripts & uv to verify tests/formatting instead of poetry.

[codecov]

Codecov Report

❌ Patch coverage is 66.66667% with 15 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
zendriver/core/connection.py	66.66%	15 Missing ⚠️

📢 Thoughts on this report? Let us know!

[github-actions]

This pull request has been marked stale because it has been open for 30 days with no activity. If there is no activity within 7 days, it will be automatically closed.

[github-actions]

This pull request was automatically closed because it has been inactive for 7 days since being marked as stale.