Improve resync performance

Resync takes a lot of time (13 minutes) with a storage that only contains 20 backups:

Machine has 4 CPUs:

I don't know why resync takes this long.

Observations

As mentioned in the comments, the resync is slow because getting the physical restore metadata takes a long time. This is mainly due to how the metadata is stored — instead of being in one file, it's split into many small files in the storage (like S3 or GCS).

For each restore, we need to read and parse all these files to figure out the status of the cluster, each replicaset, and each node. This means many small download and decode operations, which adds up quickly.

Acceptance criteria:

• make the sync/resync faster

• please specify more criteria during the implementation

• introduce a flag and config option to skip the .pbm.restore parsing. Process just the last one required for physical restore.