fix(codex): handle Responses stream whose completed event has output=null

[sree-sanak]

Problem

The ChatGPT Codex backend (https://chatgpt.com/backend-api/codex, gpt-5.x) streams the answer via response.output_item.done / response.output_text.delta events, then sends the terminal response.completed event with response.output = null.

openai-python's Responses streaming accumulator calls parse_response() on that completed event, and parse_response does:

for output in response.output:

with no None guard (observed on openai==2.24.0, openai/lib/_parsing/_responses.py). When response.output is null, this raises TypeError: 'NoneType' object is not iterable mid-stream.

run_codex_stream only catches httpx/RuntimeError, so the TypeError propagates and every request against that backend fails. Schedulers that wrap it re-surface it as RuntimeError: 'NoneType' object is not iterable.

Fix

• run_codex_stream: catch the parse_response None-output TypeError and route to the existing create(stream=True) fallback, which iterates SSE events manually and never calls parse_response. This mirrors how the code already handles the response.created/response.completed prelude/postlude quirks for this same backend.
• run_codex_create_stream_fallback and the auxiliary Codex adapter: backfill output from the collected stream items when the terminal output is None (previously only handled the empty-list case), so the already-streamed content is recovered instead of lost.

No behavior change for backends that send a well-formed response.completed.

Verification

Reproduced against the live backend (the streamed events arrive, the terminal response.completed carries output: null, and the SDK raises the TypeError). With the fix, requests that previously crashed return their content normally. Both the main conversation path and the auxiliary path (title generation, etc.) recover.

[alt-glitch]

Duplicate of #32963 (merged) — same Codex response.output=null stream recovery in codex_runtime.py and auxiliary_client.py. Root cause: #11179.

[teknium1]

Closing as obsolete. PR #33042 (merged commit cb38ce28c) removes the OpenAI SDK's client.responses.stream(...) helper from both Codex call sites entirely — we now use client.responses.create(stream=True) raw event iteration that assembles the final response from response.output_item.done events as they arrive. The terminal event's output field is never read for content reconstruction, so the SDK's TypeError: 'NoneType' object is not iterable parser crash on output=null is structurally impossible — no code path can produce it. Same defensive strategy OpenClaw uses for the same backend.

The symptom-patch + recovery logic this PR added on top of the SDK helper is now redundant. Thanks for the work during the outage.