Fix PollUtils.poll to use wall-clock time for timeout instead of accumulating sleep durations

[khandelwal-ayush]

Summary

PollUtils.poll tracks elapsed time by summing Thread.sleep(periodMs) durations instead of measuring actual wall-clock time. Time spent executing the supplier — network calls, schema registry lookups,
database queries — is not counted toward the timeout. This causes the configured timeout to inflate by a factor proportional to the ratio of supplier execution time to the sleep interval.

This was discovered during MySQL bootstrap stream creation where the connector is configured with periodMs=10ms and timeoutMs=600,000ms (10 minutes). Each schema registry call takes approximately 165ms to
fail, making each iteration ~175ms of wall-clock time while only 10ms is counted toward the timeout. After 41,184 iterations and approximately 2 hours of continuous polling, PollUtils had tracked only ~6.9
minutes of "elapsed" time, still well short of the 10-minute timeout. The actual time required to trigger the timeout is approximately 3 hours — a 17x inflation.

This affects approximately 23 production call sites across the codebase including DatastreamRestClient (12 sites covering datastream lifecycle operations), schema registry polling (7+ sites across MySQL,
Oracle, TiDB connectors), Kafka commit retries, DMS REST.li polling, and Coordinator state transitions. Any caller whose supplier performs I/O is subject to the same inflation.

The fix replaces the elapsedMs accumulator with a System.currentTimeMillis() snapshot taken at method entry. The timeout check becomes a wall-clock comparison. Both the boolean overload (used by
predicate-based callers) and the Optional overload (used by supplier-based callers) are fixed. No behavioral change for callers with near-instantaneous predicates since their wall-clock time already closely
matched the tracked elapsed time.

Testing Done

Will deploy and test

[akshayrai]

Nice. Can you include some unit tests for this?

[akshayrai]

Let's document the impact and what improvements we observe