Issue: Graceful Shutdown via InFlightTaskTracker

Summary

Enable graceful shutdown (e.g. Pod drain in Kubernetes) by tracking in-flight context tasks outside the core, subscribing to lifecycle events, and waiting for drain before exit.

Constraint / Decision

Core does not track active tasks (No Global Lock, no global registry per 9.2/10.4). Counting live in an external component (InFlightTaskTracker) in the hosting/adapter layer.

Flow

1. Hosting (e.g. IHostedService) signals intention to stop.
2. InFlightTaskTracker counts tasks by subscribing to ContextLifecycleEvent (Started / Completed).
3. On stop signal, the host waits until the counter hits zero.
4. Core ContextManager keeps working undisturbed during drain.

Scope

• InFlightTaskTracker (hosting side) counting inflight via lifecycle events.
• Integration sample with IHostedService drain + CancellationToken.
• Guarantee: no load on ContextManager; no blocking.
• Unit tests: count increments/decrements; drain completes.

Design Expectations

• Tracking lives outside Core (adapter/hosting).
• Uses lifecycle events already emitted (K9/K10) — no core changes.
• Non-blocking to ContextManager.

Non-Goals

• No policy/queue persistence.
• No APM sink.