retype failure assumes migration is needed

[John Griffith]

Currently in manager.retype there's an assumption that if the driver returns False in it's response to the retype call that it means that the backend was unable to perform the retype because it doesn't match capabilities of the type and that a migration is needed.

This is a false assumption, it's quite possible that the retype command just failed for whatever reason, if migration_policy is enabled and a migration is attempted it's possible that the the call will be made to the same backend and fail again for the same reason.

The retype response should encapsulate enough info to indicate if the backend couldn't fulfill the specs, or if it just failed. This is another case where perhaps the retype call should go through the scheduler before even getting to this point.

Code in question is here:
https://github.com/openstack/cinder/blob/master/cinder/volume/manager.py#L2126