serve: add non-streaming mode to /v1/responses; stream event parity; remove placeholder logprobs

[antznette1]

• Add stream=false support in POST /v1/responses returning a single OpenAI Response.
• Emit streaming deltas consistently; avoid placeholder logprobs by using empty lists.
• Stabilize tests: robust event checks; deterministic full request.

Files:

• src/transformers/commands/serving.py
• tests/commands/test_serving.py

Notes: Minimal diff, aligns with AGENTS.md. Improves OpenAI parity.

[Rocketknight1]

cc @gante @LysandreJik

[gante]

Thank you for the clean PR 🙏 LGTM!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[antznette1]

Hi! The docs preview link for this PR isn’t opening. I’m seeing the “INDEX doesn’t exist in pr_41353” message.