How does machine0 handle NixOS state drift and recovery in practice—for example, if a VM is manually modified outside the flake, can I detect or reset that drift, and how do snapshots interact with flake-based provisioning?