We retry requests that were sent to nodes which are unreachable or not the master,
but we also need to retry requests sent to nodes which were removed from the
cluster, since that's a different error code
Fixed by adding the appropriate exception class to correspond with the error code.
This was never a problem before, because nodes removed from the cluster used to errorneously just refuse connections instead of returning the error code. After we changed that to return the error, we need to make sure to also handle it in the client.