feat(v2): metadata index retention policy

[kolesnikovae]

The change adds a new component that is responsible for enforcement of the retention policy.

It's implemented as follows:

1. On the leader node, we periodically check which partition shards have passed the retention period threshold. The state is accessed via the Leader Read interface: it always reads the local state.
2. The component creates tombstones for the partition shards (not the blocks they contain) and issues a new raft command – truncate index.
3. The command handler removes specified partition shards from the index.
4. Tombstones then follow the standard tombstones's life-cycle: eventually, they are included into a compaction job.
5. Once the compaction job is added to the scheduler, we remove the associated tombstone object.
6. A compaction worker deletes all the objects satisfying the prefix that corresponds the partition tombstone (it includes the partition timestamp) – this part is not implemented yet.