wait readers: make monitorId required or soften rpc.md, and cover WakeSourceTracker precedence (follow-up to #4228)

Follow-up from the review of #4228 at 51f5fa8 (verdict CLEAR, two LOWs). Not blocking.

docs/rpc.md:156 says terminal monitors are named by their stable mon_ id, but extension-state.ts:88 falls back to the runtime id when entry.monitorId is missing. The field is optional on the snapshot (monitor-types.ts:59), even though every current source sets it (command-monitor-source.ts:99, file-monitor-record.ts:80). Either make monitorId required on the snapshot type or soften the doc sentence.
test/suite/regressions/3864-named-wait-readers.test.ts:84-103: the harness part of the third test goes through the binding, which already preferred items ?? monitors before feat(coding-agent): preserve named waits on the session lifecycle #4228. Only the direct wakeSourceItems assertion covers the tracker's precedence change. Add one WakeSourceTracker.observe assertion so a regression in the tracker fails a test.

Acceptance: (1) the doc and the type agree; (2) a test fails if WakeSourceTracker stops preferring items over monitors.

Related: #3864, #4228.

Two more LOWs from the review of #4231 at 107d39a (verdict CLEAR):

test/suite/extension-edit-assistant-message.test.ts:82,88 asserts a live snapshot with persisted: true and humanInputRevision: 2 while the newest durable record holds revision 1. A crash between /noop and the next input restores revision 1 and reuses revision 2. Add one line to rpc.md saying persisted does not cover a deferred admission revision.
While a custody write is failing, recompute() calls custodyWriteFailed(), which sets persisted=false, so every non-activity recompute appends a record (session-lifecycle.ts:261, lifecycle-ledger.ts:240). Limit those appends while custody is failing; the feat(coding-agent): expose live running activity in the session lifecycle #4231 fold only covered activity-only changes.

LOW from the delta review of #4243 at 25acb02 (verdict CLEAR):

In 3864-tui-lifecycle-ownership.test.ts, the new reset test checks that the extension status is gone with a bare substring of the rendered footer (not.toContain("mine")). Assert that my-ext is absent from footerDataProvider.getExtensionStatuses() instead, so both sides of the test are exact.

Navigation Menu