Cold-start agent handoffs should use the same routing semantics as warm sessions

**Motivation**
A channel-member Tech Lead should be able to hand work to an assigned specialist whether that specialist already has a live session (warm) or needs to start one (cold). In the reported workflow, a fresh Grok → Spok handoff could be delivered and initialize Spok’s agent pool, but no target turn or reply started. This makes agent collaboration depend on manually starting the specialist first and makes cold handoffs unreliable after Desktop/harness updates.

**Proposed solution**

Use the same authorization, delivery, and turn-start semantics for cold and warm agent-to-agent handoffs.
Add scoped response-routing policies: owner-only, explicit trusted agents (for example Grok/Brok), team/channel members, or a designated manager. Do not force a choice between only owner-only and unrestricted anyone.
Keep response routing separate from OS/tool permissions and approval gates.
Emit an auditable diagnostic when a handoff is delivered but no turn starts, distinguishing policy rejection, unsupported steering, pool startup, timeout, and runtime failure.
Add a regression test covering a channel-member Tech Lead starting a cold specialist session after a Desktop/harness update.

**Alternatives considered**
Just dealing with it.

**Additional context**
Anything else that helps — links, screenshots, prior art.

---

Before opening: please [search open issues and PRs](https://github.com/block/buzz/issues?q=is%3Aopen) for duplicates — link the closest one, or say "none found".