OptimizedNodeBridge cleanup can terminate workers that became busy

[bbopen]

Problem

cleanup() selects idle workers based on a snapshot (!w.busy and lastUsed) and then calls terminateWorker(worker). There’s no re-check before termination, so a worker that becomes busy after selection can be killed mid-request, causing spurious failures.

Evidence

• src/runtime/optimized-node.ts: cleanup() computes idleWorkers once, then terminates each without verifying worker.busy at termination time.
• terminateWorker() does not guard against busy state and immediately rejects pending requests.

Acceptance criteria

• Re-check worker.busy (and optionally lastUsed) immediately before termination; skip termination if it is no longer idle.
• Add a test that triggers cleanup while a worker becomes busy and verifies no in-flight request is terminated.

[bbopen]

🔄 Duplicate of #139

This issue describes the same race condition with the boolean busy flag. The fix (replacing with inFlightRequests counter) is tracked in #139 which was created as a follow-up from the PR #136 code review.