fix: clean up rebalancer diagnostic waiters

Merged
RE
3mo ago
+70-7

Follow-up to #10311 addressing valid post-merge review findings in the rebalancer diagnostic test observer.

Zero-duration waits now fault synchronously instead of depending on a thread-pool timer callback. Disposing an observer faults outstanding waiters and stops their timers so callbacks cannot outlive the observer. Timer ownership is kept directly on each waiter, with cleanup if scheduling fails.

The existing 30/60-second default waits remain unchanged and continue to rely on the test runner timeout for hangs.