Fix batch listener backoff state reset in multi-partition containers

[29shivam]

Problem

In a non-concurrent KafkaMessageListenerContainer with multiple partitions, when a successful batch follows a prior failure, doInvokeBatchListener calls commonErrorHandler.clearThreadState(). This wipes the entire FailedRecordTracker map for the current thread, resetting the failure count for all partitions — including partitions that are still inside their backoff/retry cycle.

The result: a broken record that should be retried N times (or sent to the DLQ after N attempts) restarts its counter every time any other partition successfully processes a batch, so it never exhausts its retries and never reaches the DLQ.

Reproduction scenario (from #4371):

1. Two-partition topic, non-concurrent container, backoff+DLQ configured
2. Partition A has a broken record, partition B has healthy records
3. Partition A fails → batchFailed = true, FailedRecordTracker records attempt 1 for A
4. Error handler seeks A back, commits B
5. Next poll only returns B's records → success → clearThreadState() → A's failure count wiped
6. A is un-paused; failure count is 1 again → cycle repeats, DLQ never reached

Fix

Introduce clearThreadStateFor(Collection<TopicPartition> partitions) throughout the error-handler stack:

• FailedRecordTracker — removes only the specified partitions from the per-thread failure map
• FailedRecordProcessor — public delegate to tracker
• CommonErrorHandler — new default (no-op) so existing custom handlers need no changes
• CommonMixedErrorHandler / CommonDelegatingErrorHandler — delegate to sub-handlers
• KafkaMessageListenerContainer.doInvokeBatchListener — calls clearThreadStateFor(records.partitions()) instead of clearThreadState()

This scopes the reset to only the partitions that were actually part of the successful batch, leaving backoff state intact for partitions still retrying.

The existing clearThreadState() call (on thread termination) is unchanged.

Checklist

• Tests added/updated
• @since tag added to new API (CommonErrorHandler.clearThreadStateFor)

Closes #4371