Remote firecracker snapshots

[CuriousGeorgiy]

Describe the enhancement
Rather than only using the snapshots available locally on a node, snapshots can also be transferred between nodes to potentially accelerate cold start times and reduce memory utilization, given that proper mechanisms are in place to minimize the snapshot network transfer latency. This could be done by storing snapshots in a global storage solution
such as MinIO S3, or directly distributing snapshots between compute nodes.

Problem
Currently, the blocker for using remote snapshots is container disk state restoration. Containers restored on a clean node seem to be healthy, and respond to requests, but their disk state gets corrupted after a request is received.