[ENH] `BaseClusterer` inconsistencies with other base classes

[MatthewMiddlehurst]

There are a few issues present in the BaseClusterer class IMO:

• The predict function has a y input, this is inconsistent with other collection classes and with predict_proba from the same class
• The score function is functionally different to other bases and is a required method. Can all algorithms produce scores like this? What about pipelines etc.
• fit_predict does not have a private equivalent like other classes. Also, by default it seems like this should return labels_ rather than calling _predict?