Catalog compaction. Improve minimum tx begin time propagation to replicas

After IGNITE-22641, for each catalog compaction run we have a N x M replica requests, where N - count of tables, M - count of partitions.

This looks extremely inefficient and we should consider using:

• Message coalescing.
• Depending on whether you actually need compaction coordinator or it's possible to trigger minBeginTxTime on all nodes consider sending requests to self only.
• Adding some randomization on sending requests between replication groups.