cinder schedules backups on disabled backup services

To reproduce:

1) Set up multiple cinder-backup nodes
2) Mark one of them as disabled via 'openstack volume service set --disable'
3) Create some backups
4) Note that plenty of those backups get sent to the disabled node

Why this matters:

Running 17.2, I've seen quite a few ugly lockups where a backup is scheduled, gets placed in the 'creating' state, and then stays there forever.

At least some of the time this is caused by a backup node being down but the scheduler not noticing yet; in that case the job gets scheduled on the down node and sticks forever until/if the node comes back up.

A possible (if unfortunate) workaround for this would be to mark a node as disabled (by hand) anytime it's down, but the scheduler doesn't care if a node is marked as disabled.

(Obviously the REAL issue here is the lack of any kind of feedback during the backup; ideally there would be some way to detect stuck jobs and retry but)