fix(memory-core): drain L3 cascade during pipeline destroy

[Gout999]

Summary

• Keep the destroy-time drain window open long enough for an already-accepted L2 task to enqueue its L3 persona cascade.
• Drain the L2 queue before checking L3 idle, so an L3 task created by L2 cannot be missed by destroy().
• Add a regression test that holds L3 open and verifies destroy() waits for it to finish.

Root cause

destroy() marks the pipeline as destroyed before flushing pending L2 timers. The L2 task still runs, but triggerL3() immediately returns because of the destroyed guard. In addition, _doFlush() checked L2 and L3 idle concurrently; an idle L3 check could resolve before L2 enqueued its cascade.

Fixes #1049.

Validation

• Focused Vitest: vitest run src/utils/pipeline-manager.test.ts — 1/1 passed.
• Focused TypeScript check with strict mode — passed.
• git diff --check — passed.

[Maxwell-Code07]

Thank you so much for your attention and contribution! We will arrange an internal review for this PR shortly, and all feedback will be shared right here in the discussion.