Python: fix(openai): emit image_generation_call result from streaming output_item.done

[Anish Mehta (anishmehta24)]

Motivation & Context

When the hosted image_generation tool runs with stream=True and the default partial_images=0, the Responses API delivers the final image exactly once: as a completed image_generation_call item on response.output_item.done. The streaming parser in _chat_client.py only built image contents from response.image_generation_call.partial_image events, and its response.output_item.done handler had no branch for image_generation_call, so the image was silently dropped and never even reached the "unparsed event" debug log. The response ended up with only text and reasoning, history providers persisted no reference to the image, and because responses.retrieve returns result: null for the item, the bytes were lost. The non-streaming parser handled the item correctly, so the two paths were not at parity. Affects agent-framework-openai and agent-framework-foundry.

Description & Review Guide

• Major changes: the non-streaming image_generation_call mapping is factored into a private _image_generation_item_to_contents(item) helper (same mime detection via detect_media_type_from_base64 with image/png fallback, same data: URI, same image_id and raw_representation), placed next to the existing _shell_item_to_contents. Both _parse_response_from_openai and a new image_generation_call branch in the streaming response.output_item.done handler call it, so streaming now emits image_generation_tool_call and image_generation_tool_result with the final image.
• Impact: the partial_image handler is unchanged. With partial_images > 0, callers get the preview frames flagged is_partial_image followed by the final result under the same image_id; no dedupe is done, the final render supersedes the previews. Items without a result (for example a failed generation) emit the call plus a result with outputs=None, matching the non-streaming path.
• Focus for reviewers: the four-line elif in the output_item.done handler and whether emitting on a result-less item is the desired semantics. Tests: three new tests next to test_parse_chunk_from_openai_with_file_search_call_done cover the final image, the result-less item, and streaming vs non-streaming parity via to_dict().

Prepared with AI assistance (Claude Code) and human-verified: the new tests fail on main and pass with the fix; poe fmt, poe lint, poe pyright -P openai, poe mypy -P openai and the openai test suite (511 passed, 95 skipped) were run locally.

Related Issue

Fixes #8422

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
• The PR follows the Contribution Guidelines
• This PR is linked to an issue and there is no other open PR for this issue
• This is not a breaking change.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.