* Re-trigger `ceph_access_joined` from `ceph_replication_device_changed`.
Without this, we could end up with incomplete `ceph-access` relation
if `ceph_access_joined` is executed before `ceph_replication_device_changed`.
* Seed `replication-device-secret-uuid` early in `config-changed`, to make
sure that it's set if `storage_backend` is executed before `ceph_access_joined`.
* Set `secret_uuid` as part of the `replication_device` config in the backend
config. Without this, Nova won't be able to access the proper secret for
the Cinder Ceph volumes, after a failover.
Change-Id: Ic023d05d5d17a663e1719de393bdd15f18a40484