fix: emit content_part.done and populate output_text.done.text per Responses API spec

[mikemolinet]

Summary

The /v1/responses streaming handler violates the OpenAI Responses API SSE lifecycle spec in two ways:

1. Missing event: response.content_part.done is not emitted between response.output_text.done and response.output_item.done. Per https://platform.openai.com/docs/api-reference/responses-streaming the event sequence for a text content part should be: content_part.added → output_text.delta* → output_text.done → content_part.done → output_item.done.
2. Empty text field: response.output_text.done is sent with text: "" instead of the accumulated output text. The spec requires the final accumulated content.

Accumulate the delta text at the streaming call site, emit content_part.done with the accumulated text, and populate output_text.done.text correctly.

Closes #48

Changes

• index.js: in the /v1/responses streaming handler, accumulate delta tokens in a local accumulatedText variable via accumulatedText += delta in the onChunk callback. On stream completion: (a) set response.output_text.done.text to accumulatedText (was ""), (b) emit a new response.content_part.done SSE event between output_text.done and output_item.done, with part.text set to the accumulated content. The new event is gated on partIndex > 0 (i.e. at least one delta arrived), keeping the content-part added/done lifecycle symmetric.
• index.test.js: add a local parseSseStream helper and a regression test asserting the new event, the accumulated-text content on both output_text.done and content_part.done, and the event ordering.

Testing

• npm test — 113 passed (112 existing + 1 new)
• npm run lint — clean

Notes

OpenAI Responses API streaming spec reference: https://platform.openai.com/docs/api-reference/responses-streaming.