A replicator creates an extra replica when it detects a remote disk
failure. However, when it fails to create a replica due to other
reasons (e.g. entire node failures), it doesn't create another replica
at all. We should explain it explicitly so that users can know it.
This fixes bug 906976.
Change-Id: I2f56428ccbbb0cf0d8538ca6e08f7da71257e661