fix(parsing): guard response.output None in parse_response

[Shriniwas410]

parse_response() iterates response.output directly without a None check. The Codex backend (chatgpt.com) sometimes sends response.output: null in the response.completed event even when valid output items were streamed earlier, causing TypeError: 'NoneType' object is not iterable inside the stream accumulator.

Replacing for output in response.output with for output in (response.output or []) matches the existing null-safety pattern used elsewhere in the SDK and prevents the crash without changing behavior when output is non-null.

Fixes #3325

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1eb8476f27

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Preserve streamed output when completion omits it

When a streamed response.completed event has response.output == None, this fallback parses an empty list even though ResponseStreamState has already accumulated the real output from prior response.output_item.added and delta events. In that backend case, get_final_response() and the emitted response.completed.response no longer crash, but they silently return output=[], so output_text, parsed text, and tool calls are lost despite having been streamed successfully.

Useful? React with 👍 / 👎.