fix(openai): handle null choices from model_dump() for vLLM compatibility

[Vedant Madane (VedantMadane)]

Summary

Fixes #32252

When using OpenAI-compatible APIs like vLLM, model_dump() can return null for choices even though the response object has valid choices accessible via direct attribute access. This occurs because the OpenAI SDK's Pydantic model may serialize differently when handling non-OpenAI API responses.

Changes

• Added fallback logic in _create_chat_result that extracts choices directly from the response object when model_dump() returns null
• Added unit test to verify the fix

Test plan

• Added unit test test_create_chat_result_with_null_choices_in_model_dump that simulates the vLLM scenario
• Manual testing with actual vLLM endpoint (not available in my environment)

[codspeed-hq]

Merging this PR will not alter performance

✅ 6 untouched benchmarks
⏩ 30 skipped benchmarks¹

---

_{Comparing VedantMadane:fix/vllm-null-choices-robust-handling (1f06e9b) with master (fbfe4b8)}

Footnotes

1. 30 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[Vedant Madane (VedantMadane)]

vLLM requires NVIDIA GPU. The Docker image cannot run on CPU-only systems. I had one provisioned on GCP from work when I ran into this issue but that instance has since been decommissioned after project completion.

[Vedant Madane (VedantMadane)]

Closing -- I can no longer reproduce/test this since the GCP instance with vLLM has been decommissioned. If someone else encounters null choices from model_dump() with vLLM, feel free to re-open or reference this fix.