Consider snapshotting GCP boot disks
Currently we only snapshot data disks, not boot disks, presumably under the assumption that if a machine needs to be restored it can be rebuilt and reconfigured using chef-client. This has at least the following disadvantages:
- If a machine is destroyed and recreated, there's no way to debug configuration differences shall an issue arise with the new machine
- There still exist untracked Chef changes that may not be carried over a machine recreation
- It makes our restore process dependent on completing a Chef-client run, which may not be possible under some emergency situations (e.g. the Chef server is not available)
We should consider starting snapshotting those disks