KAFKA-20423: Fix flakiness of testWakeupWithFetchDataAvailable (More)

[chickenchickenlove]

• Previous PR - KAFKA-20423: Fix flakiness of testWakeupWithFetchDataAvailable #22364

The previous PR changed the test to wait until MockClient invokes the completion callback for the fetch response. However, this callback only registers a completion handler in pendingCompletion; it does not guarantee that the fetch data has already been added to the FetchBuffer.

As a result, the following race condition can still occur:

1. The test worker finishes waiting for fetchResponseCompleted, then verifies the wakeup exception and the position.
2. After the second consumer.poll(Duration.ZERO) has passed coordinator processing, the heartbeat thread removes the handler from pendingCompletion.
3. If the heartbeat thread is descheduled before executing the handler, the test worker may observe both the completion queue and the FetchBuffer as empty and return an empty result.
4. The heartbeat thread then resumes and adds the fetch data to the buffer.

This PR changes the wait condition to fetcher.hasAvailableFetches(), ensuring that fetch data that can actually be returned is present in the buffer before calling wakeup().

This removes the dependency of the subsequent poll's record verification on the timing of the heartbeat thread's completion processing.

[chickenchickenlove]

@lianetm
Hi! Sorry for the delayed response. 🙇‍♂️
I've drafted a new PR to address the remaining flakiness in testWakeupWithFetchDataAvailable.
When you get a chance, could you take a look?