Make ngram_range two hyper-parameters in SKCountVectorizer

Suggestion.

Instead of a tuple. So that it is easier to search over their values.

But this means there will be a need for some special logic to combine this before passing to underlying sklearn.