Config server generates and persists a new ClusterId on first startup and every other node load this Cluster ID from the config server at startup using the ClusterIdentityLoader .

The problem is that currently in case no config version document is found on the config server at startup the ClusterIdentityLoader will load a fake zero-filled OID.
This is wrong and can bring to unexpected error when we later try to add a shard.

Instead we should make ClusterIdentityLoader::loadClusterId() to simply fail if no config version document is found on the config server.