RootflowDataset split should support splitting out a test set as well.

[HesitantlyHuman]

The datasets should support test sets in addition to training and validation. Adding a parameter like test_proportion=None should be sufficient.