Spike: snapshotting clusters to speed up testing

[weyfonk]

Fleet's end-to-end tests often suffer from flakiness caused by leftover resources, or timeouts caused by different specs between developers' machines and Github runners. Test suites could run faster and more reliably on clean clusters, if those could be spun up and torn down faster than they currently are. Let's look into our options:

• etcd snapshotting
• using k3d to snapshot containers; possibly related issues/discussions:
  • https://github.com/k3d-io/k3d/issues/160
  • https://github.com/k3d-io/k3d/discussions/1244 → mentions SQLite being the default datastore
• other?