Include text/event-stream header only when stream=True

[vladimirivic]

Summary:
We want to use the headers to negotiate content.

Sending this header in every request will cause server to return chunks, even without the stream=True param.

llama-stack-client inference chat-completion --message="Hello there"

{"event":{"event_type":"start","delta":"Hello"}}

{"event":{"event_type":"progress","delta":"!"}}

{"event":{"event_type":"progress","delta":" How"}}

{"event":{"event_type":"progress","delta":" are"}}

{"event":{"event_type":"progress","delta":" you"}}

{"event":{"event_type":"progress","delta":" today"}}

Test Plan:

pip install .

llama-stack-client configure --endpoint={endpoint} --api-key={api-key}

llama-stack-client inference chat-completion --message="Hello there"
ChatCompletionResponse(completion_message=CompletionMessage(content='Hello! How can I assist you today?', role='assistant', stop_reason='end_of_turn', tool_calls=[]), logprobs=None)

[ashwinb]

this should be if stream, but the higher level issue is that this is generated code. We need to make sure we auto-apply this patch always after generation (see stainless_sync.sh) or find another way

[yanxi0830]

Is this still needed after #108 ?

[ashwinb]

@yanxi0830 yes this is still needed but for a different reason. this header is sent by the client to the server, not the other way round.