Consider every ClickHouse operation as a merge

[nikitamikhaylov]

Use case

To simplify the code, especially the part deletion process we can say that part deletion is a merge to an empty part which will create am empty covering part (and it would be a part of replication process).

The current solution is complicated and we still find some race-conditions. For example the one fixed recently: #61610

The flip side of this mechanism is that ClickHouse will always leave an empty part in old partitions - similar to the block_number nodes in Keeper. To overcome it we can reuse SYSTEM FORGET PARTITION query introduced in #58925

WDYT? @den-crane

[den-crane]

I don't foresee problems if SYSTEM FORGET PARTITION will be implemented.

I am NOT sure how it will impact users who move data to disks with prefer_not_merge, but such merges can disobey prefer_not_merge or you can eridicate prefer_not_merge.