LCM.fit_transform does not follow sklearn philosophy

[remiadon]

Describe the workflow you want to enable

For now skmine.itemsets.LCM implements a fit_transform methods, returning a DataFrame
with two columns, namely itemsets and supports
To be fully compatible with sklearn this is not sufficient

Describe your proposed solution

move what the current fit_transform function is doing to a new function ("discover ?"), so we can still get a pretty dataframe with itemsets and supports,
and change the fit_transform method, so that:

• it returns a binary pandas.DataFrame, with itemsets as column
• the width of this DataFrame is constant and pre-defined (pass n_features as arg ?)

Additional context

Ideally we should be able to use our pattern mining algorithm to extract feature from transactional
datasets.
see this repo