Stream a Perplexity Sonar chat completion token-by-token while still reliably capturing citations and search_results.

Documented steps

1. Set "stream": true in the chat completions request body (https://docs.perplexity.ai/docs/sonar/features).
2. Consume the response as server-sent chunks; read partial text from chunk.choices[0].delta.content, the same shape as OpenAI streaming.
3. Do not expect chunk.search_results or citations on early chunks — the docs state search results and metadata are delivered in the final chunk(s) of a streaming response, not progressively.
4. Accumulate delta.content client-side for the visible answer, and separately watch for chunk.search_results and chunk.usage arriving at the end of the stream.
5. Once the final chunk arrives, merge its search_results/usage into your accumulated answer before showing citations to a user or downstream agent.
6. If you need citations to render progressively as the model writes them, this Sonar SSE shape won't provide that — it only guarantees them at the end (https://docs.perplexity.ai/docs/sonar/features).