fix(queue): waitForDrain poll fallback for drains that never emit

[ldt1996]

Fixes #1642

• waitForDrain() now arms a 100ms poll alongside the drained listener, so a waiter can no longer hang when the queue empties through a path that never emits the event (the on('data') attach loop, the resolveNext bypass). Regression test included.
• Also exposes Table.evictionMS beside expirationMS; an upcoming harper-pro change needs to detect evicting tables.

Lavinia, via Claude

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request introduces a polling fallback mechanism in IterableEventQueue to prevent waitForDrain from hanging when the queue empties through paths that do not emit the drained event, accompanied by a new unit test. It also adds a static evictionMS getter in Table.ts. Feedback was provided regarding a potential memory leak in IterableEventQueue.ts where the once('drained') event listener is not cleaned up if the queue is drained via the polling fallback or closed; a code suggestion was provided to clean up the listener in the settle function.

[claude]

Reviewed; no blockers found.

[github-actions]

Patch cherry-pick: conflict

Cherry-pick onto v5.1 produced conflicts on commit(s): fbc40f088c39c48612a86866241bad2459ad69eb

The conflict markers are committed on branch cherry-pick/v5.1/pr-1643.
A pull request has been opened to land this patch: #1692

[ldt1996]

Good catch, fixed in bb63b78: settle now removes the drained listener, and the test asserts listenerCount goes back to zero after a poll-path settle.
If the unit check goes red again it should be the same LMDB flake that's currently failing on main

Lavinia, via Claude

[kriszyp]

Reviewed via Claude review-queue — solid poll-fallback fix for drains that never emit; unref'd 100ms poll only while waiting, negligible cost. Unrelated Windows CI flake persists (also red on main). LGTM.