fix(coding-agent): preserve sessions during daemon startup

[traditio]

Summary

• start the saved-session catalog through a dedicated lightweight entrypoint and allow slow filesystems a 30-second startup window
• treat a connected supervisor without a hello as still starting instead of stale, and revalidate the connected daemon before any stale shutdown
• start the catalog before terminating a recoverable worker so an interruption marker can be persisted before replacement

This prevents a slow cold start from cascading into supervisor churn, failed worker recovery, and Unknown active session errors while the durable session is still on disk.

Validation

• npx tsx ../../node_modules/vitest/dist/cli.js --run test/daemon-catalog-process.test.ts test/daemon-launch.test.ts — 32 tests passed
• npm run check — passed (formatting, type checking, installer render, browser smoke)
• manually exercised DaemonCatalogClient.start() on the affected WSL-mounted checkout; the dedicated entrypoint reached ready consistently below the former 5-second cutoff

Note

Fix session loss during slow daemon startup by waiting for catalog readiness and revalidating supervisors

• probeDaemonVersion in daemon-launch.ts now distinguishes 'unresponsive' daemons (connected but no hello within timeout) from 'stale' ones, and ensureDaemonRunning waits within the startup window before giving up.
• shutdownStaleDaemonIfNotBusy returns a tri-state ('current' | 'stopped' | 'busy') and skips shutdown if the daemon sends a current hello during staleness checks, preventing replacement of a healthy daemon that was slow to start.
• DaemonSupervisor.recoverUncertainWorkerOperations in daemon-supervisor.ts now awaits catalog.start() before sending SIGKILL, ensuring the catalog is ready to record interrupted work.
• The catalog process in daemon-catalog-process.ts now has a 30-second startup timeout and launches via a dedicated entrypoint under Node environments.

^{Macroscope summarized e6407a4.}