Add clustering methods to the test learners

[ablaom]

Include one example of that generalizes to new data, and one that doesn't, and implement predict(learner, ::LabelAmbiguous, X) ("fit" and "predict" in one call) in both cases to give a common predict interface, and show this in the doc-strings.