…nterrupt offset translation

MirrorCheckpointTask reloads the last checkpoint at start,
OffsetSyncStore stores OffsetSyncs before reading till end.

Add test case simulating restarted task where the store is reinitialized
with later OffsetSyncs and check that emitted Checkpoint do not rewind.

Also addresses KAFKA-16622 Mirromaker2 first Checkpoint not emitted until
consumer group fully catches up once because the OffsetSyncStore store
is populated before reading to log end.

Co-Authored-By: Adrian Preston <prestona@uk.ibm.com>

Co-Authored-By: Adrioan Preston <prestona@uk.ibm.com>

use previous OffsetSyncStore "pessimistic" load logic

Co-Authored-By: Adrian Preston <prestona@uk.ibm.com>

Use a wrapper to the checkpointsPerGroup map to ensure thread safety on restart.

The callback rethrows the error as a way to stop the KafkaBasedLog looping forever in readtoEnd.
This occurs when unauthorized to READ from the topic, as KafkaBasedLog retries indefinitely.
Unauthorized to DESCRIBE instead fails immediately before fetching.

All such failures result in a warning log message about degraded offset translation.

for consistency with OffsetSyncStore

CheckpointStore started debug/trace message