Fix NameError in MondrianCategorizer.apply() when oob=False

[shivamlalakiya]

MondrianCategorizer.apply()'s oob=False branch reads a bare name learner,
which is neither a parameter of apply() nor an attribute on the instance —
it is self.learner, assigned in fit(). The branch therefore raises
NameError on every call:

from crepes.extras import MondrianCategorizer
from sklearn.ensemble import RandomForestRegressor

rf = RandomForestRegressor(n_estimators=10, oob_score=True, random_state=0).fit(X, y)
mc = MondrianCategorizer()
mc.fit(X=X, learner=rf, oob=False)   # succeeds
mc.apply(X)                          # NameError: name 'learner' is not defined

oob=False is fit()'s documented default (its own docstring documents the
learner parameter), so this is the documented, default combination, not an
edge case. The oob=True branch two lines above already reads
self.learner.estimators_, so this is the one call site that never picked up
self..

One-line fix: learner.predict(X) → self.learner.predict(X). The branch
currently raises unconditionally, so no existing output can change.