Test coverage: mailbox owner-resolution wiring (escalateHeldToOwner) and cleanupBuilder() held-row dismissal

[mohidmakhdoomi]

What

Two pieces of the Spec 1313 maintainer round (starvation-alarm + cleanup) are verified today by inspection and reuse of already-tested resolvers, but have no dedicated test. Add direct coverage for both:

1. Owner-resolution wiring (escalateHeldToOwner). Exercise the live resolution path — architect-skip (never notify an architect about held mail), the spawning-architect → main → first-registered fallback, and the supersede-keyed enqueue (one coalesced notice, no notice-about-notice) — via the exported makeDeliveryPorts seam against a seeded global.db registry.
2. cleanupBuilder() held-row dismissal. Assert that afx cleanup transitions a removed agent's held rows to dismissed (via dismissHeldForAgent) so they stop pinning heldCount / escalated state forever.

Why

These are the two glue points where the round-3 primitives meet the live system, and they're exactly the parts a refactor could silently break without a test noticing — the resolvers underneath (resolveAgentInRegistry) and the drainer→port contract are already covered, but the wiring that invokes them from the escalation drainer and from cleanupBuilder is not. The maintainer flagged this as an optional fast-follow (optional-2) rather than a merge blocker.

Notes

• The gap is real but low-risk: resolveAgentInRegistry and the round-3 drainer→port contract are tested; this covers the invocation/wiring layer on top.
• Cost is why it was deferred: the owner-resolution test needs global.db registry seeding and the cleanup test needs git-worktree mocking — disproportionate to fold into the delivery PR, appropriate as a standalone.
• Refs afx send: mailbox-first delivery — persist, gate on rendered-empty, never force-inject #1313 (PR [Spec 1313] afx send: mailbox-first delivery (never force-inject) #1330), Follow-up section of codev/reviews/1313-afx-send-mailbox-first-delivery.md (maintainer optional-2).