rebase: adapt to vLLM v0.29 APIs

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>

tests/entrypoints/test_stream_finish_reason.py

@@ -411,7 +411,7 @@ async def test_audio_chunk_without_waveform_keeps_stream_alive():
 @pytest.mark.asyncio
 async def test_audio_choice_error_response_is_not_iterated_as_choices():
     """Any ErrorResponse from the audio path is skipped, not iterated."""
-    from vllm.entrypoints.openai.engine.protocol import ErrorResponse
+    from vllm.entrypoints.serve.engine.protocol import ErrorResponse
 
     serving_chat = build_serving_chat()
     request = make_request(modalities=["text", "audio"])