Pass the keyring over the wire as part of response of initialize_connection() call instead of implicit assumption that rings are already present

[Jyoti Ranjan]

The bug is in continuation of issue seen at https://review.openstack.org/#/c/351005/. The suggestion was to decouple cinder backup service with cinder volume service deployment. As of now,
cinder backup service assumes that keyring for ceph for volume operation is present on node running cinder backup service instance. It is an implicit assumption and expected to be
fullfilled by cloud deployer. We can break this hard coupling in two ways:

1. [Medium term] Pass the keyring over the wire as part of response of initialize_connection() call.
2. [short term] Relying on barbican to manage barbican keys association of cinder-volume and cinder-backup services. Cons: cinder will have dependency on barbican

This defect is to address the coupling aspect in medium term.

Related defect:

https://review.openstack.org/#/c/351005/: tells about why we need to decouple cinder backup and cinder volume service