fix(monitor): drop stale completion wakes

[richardhristov]

Summary

• guard the central loop:fire emission against stale extension contexts
• drop an obsolete wake instead of crashing Pi when monitor completion races session replacement or reload
• cover the direct MonitorCreate onDone completion path with a regression test

Root cause

#33 guarded events emitted by MonitorManager and awaited monitor shutdown during session teardown. However, an onDone completion callback calls onLoopFire() directly, and emitLoopFire() still used an unguarded pi.events.emit(). If that callback reaches the old extension runtime after it becomes stale, the synchronous emit throws as an uncaught exception.

Non-stale event-bus errors continue to propagate.

Verification

• npm run typecheck
• npm run lint (passes with 4 existing optional-chain warnings)
• npm test (42 files, 658 tests)
• npm run build
• npm run test:package
• npm audit --audit-level=moderate

[richardhristov]

@trvon This is a regression in the stale-context path addressed by #33: direct onDone completion callbacks bypass MonitorManager.emit() and reach the raw loop:fire emit. CI is currently awaiting maintainer approval because this PR comes from a fork. Could you approve the run and review when convenient?

Dismissing review and continuing locally

[trvon]

Resolved in two signed maintainer commits:

• 6d859e4 adds a RED behavioral regression proving a stale runtime consumed and deleted an active loop's final fire without delivering a wake.
• 7e83767 probes extension-context liveness before any fire accounting, retirement, auto-task creation, monitor-loop deletion, or workflow-wait mutation. The monitor-specific regression now also verifies that stale completion never reaches loop:fire and leaves its loop intact.

[trvon]

@richardhristov Thanks for the PR!