fix(codex): monkey-patch openai SDK parse_response to handle output=None

[nicha16]

The OpenAI SDK's parse_response can return None for certain edge cases (non-recoverable stream errors, cancellation mid-parse). Without handling, this propagates as AttributeError: 'NoneType' object has no attribute 'model_dump'.

Adds a monkey-patch in codex_responses_adapter.py that wraps parse_response to return an empty completion instead of None.

[alt-glitch]

This was generated by AI during triage.

Duplicate of #32963 (merged 2026-05-27, "recover Codex Responses streams with null output"). That fix — plus the structural refactor in #33042 — already handles the Codex output=None case in agent/codex_responses_adapter.py; main no longer needs an OpenAI-SDK parse_response monkey-patch. Four earlier PRs taking this same monkey-patch approach (#33173, #32959, #33275, #33553) were closed as duplicates for the same reason.

[teknium1]

Closing as redundant against current main.

The Codex Responses output=None crash this targets is already fixed by the event-driven rewrite of agent/codex_runtime.py (merged as #32963 and hardened afterward): the main agent loop and the auxiliary client both consume the raw SSE stream via responses.create(stream=True) and assemble the final response from response.output_item.done events, rather than going through the high-level responses.stream(...) / SDK parse_response helper that reads response.completed.response.output. Since we never read that field for content reconstruction, null output can no longer crash us — so monkey-patching the SDK's parse_response is no longer needed on either path.

Thanks for the fix — the approach was sound; the code path it patched just no longer exists on main.