Public thread-summarization API

[zoharbabin]

Why

There's no public API for getting a summary of a thread. Backend investigation confirms summarization logic already exists server-side in more than one form (used internally at different points in a conversation's lifecycle), but none of it is exposed as something an integrator can call directly. Anyone wanting "give me a summary of this conversation" today has to build it themselves by replaying the transcript into a separate prompt.

What

A public Conversations/threads method to request a summary of a given thread, backed by the summarization capability that already exists server-side.

How

• This requires the backend to expose an existing internal capability through a public endpoint — the summarization logic itself doesn't need to be built from scratch, only surfaced.
• Once available, add a thin SDK client method (e.g. Conversations.summarize(threadId)), following the same shape/error-handling conventions as the existing Conversations methods.
• Document it in API-REFERENCE.md with a worked example.

Success criteria

• Calling the new method against a live thread with several turns returns a coherent summary, verified live.
• The method's error semantics match the rest of Conversations (same non-2xx handling, no silent empty-result cases).

Where it lives

Core — ./management (Conversations). Depends on a backend endpoint that doesn't exist publicly today.

Priority / Labels

Low-medium. Useful but not blocking; existing internal summarization already covers the main use cases indirectly.

---

Part of #46.