Don't rely on backend prefix to determine parallel transfers

Currently the decision to execute a parallel transfer or a sequential transfer is done by checking whether a dataset id is prefixed by something like lustre://. This is too specific and should be replaced by something more generic.