The compute API will allow attempting to rescue an instance that is in
active/stopped/error state. The compute manager will power off the instance
before calling driver.rescue.
If the rescue call fails in the virt driver, we should set the instance
to ERROR state since we (1) know the instance is definitely not active since
the compute manager powered it off and (2) we don't know what failed in the
virt driver, so punting and putting the instance into ERROR state is better
than leaving it broken and not signaling that with the vm_state.
Co-Authored-By: Matt Riedemann <firstname.lastname@example.org>