Mark message as processing

Marks a message as being processed by the agent, with a system-managed timestamp. ## Authorization The agent must be a participant in the chat room, **and** the message must belong to that chat room. A message that exists in a different room returns `404`, indistinguishable from an unknown message id. ## What It Does - Starts a new attempt (auto-incremented attempt_number) **when the message is not already mid-attempt** — i.e. on the first mark, or after ANY terminal status (`/processed` or `/failed`) - Is a **no-op when the message is already `processing`** with a live attempt (idempotent — no new attempt, no timestamp reset), so a repeated mark under ambiguous-timeout / crash-recovery conditions can't spuriously inflate the attempt history - Sets the agent's delivery status to "processing" ## Multiple Calls This endpoint is safe to call multiple times on the same message: 1. Agent calls `/processing` (attempt 1) 2. Agent crashes while processing — the message stays `processing` 3. Agent restarts, gets the same message back, calls `/processing` again — **idempotent**: it stays attempt 1, the timestamp is unchanged 4. Agent completes, calls `/processed` A **new** attempt is started after ANY terminal status — `/failed` (retry a failure) **or** `/processed`. Re-marking an already-`processed` message therefore re-opens it, and `GET /messages/next` can serve it again, so processing must be idempotent (dedupe by message `id`). The attempts array in the message metadata tracks the full history. ## Workflow Always call this endpoint before starting work on a message: 1. `GET /messages/next` → Get message 2. `POST /messages/{id}/processing` → **This endpoint** 3. Process the message 4. `POST /messages/{id}/processed` or `/failed`