Improve engine-level object listing for policer

[roman-khimov]

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

I'm always frustrated when I think of how policer picks up objects for its job. It effectively works shard by shard, which means that duplicated objects are tried several times and we never know about it.

Describe the solution you'd like

Pick N objects from all shards, then deduplicate them, then return to policer with additional per-shard data, this allows to:

• not try broadcasted (lock/ts/link) objects several times
• have additional data about local placements to leverage it for After replication of object to another shard, the object remains on the returned disk forever #2087

Describe alternatives you've considered

Things work as is as well. Notice that the new scheme will also require a bit more resources (always checking all shards).

[roman-khimov]

Head/HeadECPart calls can be eliminated in this case as well.