0.2.0: Handle update-reboot failures/ "crash loops"

[cbgbt]

If an update-reboot fails and is forced to rollback to the previous partition, brupop will continue to attempt to add the node to the active set, updating it endlessly and needlessly.

We should record attempted updates (or more generically, attempted state transitions) and give up after some number of attempts, recording a failure metric.

[cbgbt]

This is merged, but we want to use the migrations work in #161 to ensure a migration exists from existing 0.2.0 CRD. We'll release this in another pre-GA release.