[Forknet] State parts GC

[VanBarbascu]

Due to the short epoch length, the state parts bucket for a forknet can get quite large if the network is not restarted.

This task has a lower priority as we plan to implement full descentralized state sync.

Possible option:

• Set retention period on the bucket. This has the drawback of not taking into account the time when the network is parked. When the network is resumed, the parts will not be available and the dumper need to be configured to always check if the parts are dumped in the bucket. (state_sync.dump.restart_dump_for_shards)
• Set a cron job on the dumper node to clear all old epochs in the bucket.