fix(core): retry failed session wakes

[kitlangton]

Summary

• retry a failed advisory Session wake once without requiring another external message
• prefer newer coalesced work over retrying an older failed wake
• hide a recovered first failure from awaitIdle
• report persistent advisory failures once after the bounded retry is exhausted

Why

OpenCord observed durably accepted Discord follow-ups becoming dormant after a transient Session drain failure. The coordinator removed the final advisory lane and only logged the failure, so work resumed only after another human message triggered a new wake.

Durable terminal failure events are intentionally split into a stacked follow-up.

Verification

• bun run test -- test/session-run-coordinator.test.ts test/session-runner.test.ts
• bun typecheck in packages/core
• repository pre-push typecheck: 22/22 tasks passed