Fix /v1/responses streaming: emit full OpenAI Responses API event sequence

[121212165]

Problem

When calling POST /v1/responses with "stream": true, the server only emits three events: response.created -> response.output_text.done -> response.completed.

Clients built on the official OpenAI SDK (and most third-party clients) parse the Responses API stream with a strict state machine. They require response.output_item.added and response.content_part.added before any text event, and render text from response.output_text.delta. Without these events the whole output is discarded and the user sees an empty reply ("no content returned"), even though the HTTP status is 200.

Fix

Emit the complete standard event sequence:

response.created
response.in_progress
response.output_item.added
response.content_part.added
response.output_text.delta
response.output_text.done
response.content_part.done
response.output_item.done
response.completed

• Added sequence_number and output_index fields to every event
• function_call items also get output_item.added / function_call_arguments.delta / output_item.done
• Non-streaming path unchanged

Testing

Verified locally on Windows (Python 3.12):

• stream: false unchanged, returns completed response
• stream: true now emits all 9 events in order; a client that previously showed "no content" renders replies correctly after this fix