v0.4.0

Notes

Adds thread management and multi-turn conversations, and makes rate-limit errors
actually readable. Builds on the v1 redesign (0.3.0).

⚠️ Breaking changes

• user_id moved to the client and is now mandatory. Set it once via
  AsyncCominty(user_id="user_...") or COMINTY_USER_ID; it's validated at
  construction and applied to every request. It's been removed from
  chat.start (and all other call signatures).

  # before (0.3.0)
  await client.chat.start(agent_id=..., message=..., user_id="user_...")
  # now (0.4.0)
  client = AsyncCominty(user_id="user_...")
  await client.chat.start(agent_id=..., message=...)

✨ New

• client.threads : full thread management: list (scoped to the user, with
  terms/limit/page), get, update (partial), archive.
• chat.send(thread_id, ...) : follow-up messages in an existing thread; same
  streamable run as start. Use it to answer an agent's clarifying questions.
• AssistantRun.questions() : read the agent's clarifying questions
  (prompt + options) when it needs input instead of answering.
• Clearer 429s. RateLimitError now renders a scope-aware, actionable message
  (organization quota vs. user quota vs. concurrency cap) and exposes
  .scope, .retry_after, and .reset_at.

🐛 Fixes

• chat.send now correctly parses POST /chat/{id} as the assistant Message
  (it returns a message, not a thread).
• threads.update now correctly returns a ThreadSummary (PUT /chat/{id}
  responds without the message history).
• Removed a stray pdb.set_trace() from the transport layer.

📚 Docs & examples

• Eight runnable examples under examples/:
  streaming, await-result, follow-up, answering questions, listing/managing
  threads, calling a custom managed agent, and pulling live context from an
  MCP server (Linear) with colored, aligned rich output.
• New SDK & API reference at docs/sdk-reference.md; README rewritten to the
  real surface.