When localization fails due to disk attached RW to another instance, check if the other instance is stopped

[dheiman]

and if it is, delete the other instance and retry. This will solve itinerant issues with localizer VMs not self-destructing when done.

Alternatively (or additionally to be extra robust), ensure disks are detached at the end of localization.