KafkaSinkCluster: use leader_epoch

[rukai]

We need to change our metadata response processing to store each partitions leader_epoch and make use of it when updating our partitions metadata state.
We should only update a partition when the leader epoch in the response is newer (a smaller integer) than the existing epoch. This ensures that if we get the metadata from a node that is a bit out of date we dont regress the state of the metadata.
However in order to enable cleaning up of deleted partitions we should still delete partitions if missing in the response, since otherwise there would be no way to clean these up.

This PR implements this.

[codspeed-hq]

CodSpeed Performance Report

Merging #1660 will not alter performance

_{Comparing rukai:leader_epoch (00635f5) with main (9f89390)}

Summary

✅ 37 untouched benchmarks