Returning completely from the upgrade procedure when
any hosts at all don't have their metadata up-to-date
is unnecessarily strict. We can just skip daemons
specifically on hosts where the metadata is not up-to-date
and then just make sure the metadata is up-to-date on all
hosts before calling the upgrade complete
Fixes: https://tracker.ceph.com/issues/53598 Signed-off-by: Adam King <adking@redhat.com>