ebucketmigrator mis-handles situation when all vbuckets to replicate are not ready

SUBJ. We had this code, but lots of recent changes in this area apparently removed this. No vbuckets ready may unfortunately (but supposedly very unlikely) happen when after vbucket move we find that some replicas are later w.r.t. checkpoints than replica before them. Which will cause backfill into those later replicas, which will cause replication from them to be impossible until backfill is done. Our usual way of handling them is by not replicating such not ready vbuckets and checking periodically if they have became ready. As pointed above we don't deal correctly with case when all vbuckets we need to replicate between pair of nodes are not ready. In such case we currently pass empty vbucket filter which causes ep-engine to send all vbuckets instead of none.