Update how plm.predictor takes in the test data

Currently, plm.predict takes a set of features and labels as test data.

But it would be better if it could, analogously to plm.trainer, take the list returned by data.splitter as input