fix(codex): handle None response.output from SDK get_final_response (fixes #32908)

[Morad37]

Summary

When the Codex backend omits the response.completed terminal event, the OpenAI SDK's parse_response() crashes with TypeError: 'NoneType' object is not iterable at line 61 of _parsing/_responses.py - because response.output is None instead of []. This TypeError bypasses the existing backfill logic at codex_runtime.py:250-273 designed exactly for this scenario.

Fix

Catch TypeError on stream.get_final_response() and set output=[] on a minimal response object. The existing backfill then reconstructs output from already-collected stream items (collected_output_items) or synthesized text deltas - preserving data and avoiding a fresh API call (no double-billing).

The existing exception handler at line 302 caught RuntimeError but not TypeError, so the crash jumped past every recovery path and escalated as a non-retryable client error.

Testing

Bug is reproducible with hermes chat -q "say the word ok" --provider openai-codex -m gpt-5.5 (45+ times per session per the reporter).

Closes #32908

[alt-glitch]

Duplicate of #32884. Same Codex null output recovery. Competing with #32888, #32890, #32891, #32897, #32898, #32900, #32901, #32913, #32914, #32915, #32918, #32920. Root cause: #11179.

[teknium1]

Closing as duplicate — the Codex null-output fix has been merged via #32963 (cherry-picked from @carltonawong's PR #32890). Thanks for the help during the outage. Closes #11179.