Detect loss of metadata store disks

[tillrohrmann]

Once a metadata store peer is part of a cluster and loses its disk, it should not join the cluster with the same peer id in an already existing configuration epoch. The problem is that this peer might have cast some votes before on which it must not go back. Hence, if we can detect this situation, then we can prevent this node from joining the cluster again. One idea could be to store in the NodesConfiguration whether a node has joined the metadata store cluster or not. If it did and we find an empty disk, then this is an indicator that we lost the disk.