Shard data balancing

[roman-khimov]

Is your feature request related to a problem? Please describe.

I'm always frustrated when disk replacements leave nodes in unbalanced state, new disks store less data than they should. Same thing happens if extending node with new shards, old shards keep old data and never share it with new ones. This makes the load uneven and breaks redundancy assumptions since loss of one drive is not equal to loss of another in terms of lost object count (even though the basic redundancy unit is SN we can assume that shards are equal).

Describe the solution you'd like

Policer has a clear picture of local data balance since #3862 (which is leveraged by #3908), the same data can be used to trigger PUT into more appropriate shard and DELETE from inappropriate one.

Describe alternatives you've considered

This has to be solved anyway.