fix(codex): combine Responses timeout fixes

[wysie]

Summary

Backports and combines two Codex Responses timeout fixes onto current main:

• References fix(codex): account for Responses payload timeouts #24126: preserve configured request timeouts through the Codex Responses request path.
• References fix(codex): estimate stale timeout from Responses payload #31855: estimate non-stream stale timeouts from the full Responses payload instead of only Chat Completions messages.

This also resolves the current helper-based refactor conflict by applying the stale-timeout logic in agent/chat_completion_helpers.py, while keeping run_agent.py as a forwarder. It preserves xAI Responses extra_body preflight support while adding timeout to the allowlist.

Test plan

• git diff --check
• .venv/bin/python -m compileall -q run_agent.py agent/transports/codex.py agent/codex_responses_adapter.py agent/chat_completion_helpers.py
• HERMES_HOME=/private/tmp/hermes-agent-test-home venv/bin/python -m pytest tests/agent/test_non_stream_stale_timeout.py tests/agent/transports/test_codex_transport.py tests/run_agent/test_run_agent_codex_responses.py tests/run_agent/test_run_agent.py::TestBuildApiKwargs -q -o addopts=

Result: 140 passed, 1 warning.

[teknium1]

Thanks @wysie — closing this as redundant. The Responses-payload stale-timeout estimation + timeout propagation through the Codex transport/preflight landed on main earlier today via #29507's follow-up (commit 2d42272 by @kasunvinod), which covers the same two upstreams (#24126 and #31855) you backported here. Your work helped confirm the design direction.

The remaining gap — first-byte detection on the Codex stream itself — is being salvaged from #31984 in #32042. Appreciate you helping triage this.