add transform to do model stacking

[sdpython]

model stacking is the possibility to use output of a predictor as feature the next transform or learner in a pipeline.