KAFKA-20091: Fix inconsistency in time-based retention checks between remote and local segment deletion logic.

[jiafu1115]

When currentTime- largestTimestamp of segment = retentionMs, the behavior is different:

1. Local Segment delete: Not delete

org.apache.kafka.storage.internals.log.UnifiedLog#deleteRetentionMsBreachedSegments   
boolean delete = startMs - segment.largestTimestamp() > retentionMs;

2. Remote segement delete: Delete

org.apache.kafka.server.log.remote.storage.RemoteLogManager.RLMExpirationTask.RemoteLogRetentionHandler#isSegmentBreachedByRetentionTime
shouldDeleteSegment = metadata.maxTimestampMs() <= retentionTimeData.get().cleanupUntilMs;

//cleanupUntilMs is time.milliseconds() - retentionMs;
private Optional<RetentionTimeData> buildRetentionTimeData(long retentionMs) {
    long cleanupUntilMs = time.milliseconds() - retentionMs;
    return retentionMs > -1 && cleanupUntilMs >= 0
            ? Optional.of(new RetentionTimeData(retentionMs, cleanupUntilMs))
            : Optional.empty();
}

cc @kamalcph

Thanks for your comments on KIP-1241 .
While I thinking through one of the your comments and reading the code, I noticed a potential problem in this area, so I opened this PR. Many thanks.