feat(agents): make agent create/role-edit approval configurable — mandatory owner confirmation caps team-building velocity

**Motivation**

When an owner runs a multi-agent team (e.g. a manager agent coordinating specialist agents on a project), onboarding a new agent or adjusting an existing agent's role definition is routine work. Today both `agents draft-create` and `agents draft-update` always land as owner-reviewed drafts that only the owner can approve and save from Buzz Desktop. The owner becomes a mandatory serial gate on every team-management operation:

- Real case (2026-08-18): a multi-agent project team coordinated by a manager agent needed to onboard an additional specialist and adjust one role definition mid-project. Both changes stalled waiting for owner confirmation; work on the affected track resumed only after the owner manually approved in Desktop. This round-trip delay is the bottleneck, not the edits themselves.
- Updating a role definition (e.g. a system-prompt tweak) carries the same full owner-review weight as creating a brand-new agent, even though it is a routine edit.
- Managed agents cannot drive even the draft step from their own runtime: `buzz agents draft-create` / `draft-update` require the owner's NIP-OA auth tag, which managed agents do not hold. The whole loop is owner-in-the-loop by construction.

Net effect: team-building velocity is capped by owner availability, which undercuts the "hive mind / human-agent workspace" positioning — the owner delegates day-to-day work to agents but must personally babysit the team's own composition.

**Proposed solution**

Make the owner-review gate a configurable trust policy rather than a hard requirement:

1. **Trust tiers per workspace (or per agent)**:
- `owner-only` — current behavior (safe default)
- `manager-delegated` — the owner designates one or more manager agents; they may approve create/edit drafts for agents in the subtree they manage, with every approval recorded in the audit log
- `open` — any workspace agent may create/edit agents; all changes audited, owner receives a digest notification
2. **Batch approval UX in Desktop** (lighter interim step): review-and-save N pending drafts in one pass instead of one round-trip per draft.
3. When the actor is already trusted for the target agent, save immediately and surface "auto-approved by \" in the audit log.

**Alternatives considered**

- Keep mandatory owner review for everything (status quo) — rejected: it serializes routine team operations on human availability; multi-agent teams cannot self-organize even inside the trust boundary the owner already established.
- Time-based auto-approval (draft activates after N minutes) — weaker: silent activation of role changes could surprise the owner; explicit policy beats implicit timeout.
- Hand the owner's NIP-OA auth tag to agents so they can approve their own drafts — works today but is all-or-nothing; there is no scoped delegation primitive yet. Scoped managed-agent signing (#5381) is the mechanism this feature should build on.

**Additional context**

Closest existing issues (searched 2026-08-19):
- #3791 — bug: draft-create succeeds on relay but does not trigger the Desktop review form; fixing that UX bug reduces friction but does not remove the gate.
- #5129 — owner-reviewed draft-create via workflows: this request is about making that gate configurable, not removing it.
- #5381 — owner-approved scoped managed-agent signing: the scoping primitive this feature would leverage.
- #5372 — multi-agent routing/wakeups/observability: same theme, agent teams operating with less human-in-the-loop.
- #3227 — headless owner-approved app-integration agents: adjacent approval UX for a different agent kind.