Update the StackingEstimator to ease feature names retrieval and understanding.

While I was using TPOT in a interpretable ML project, I suffered to recover the resulting feature matrix of the transformation from the StackingEstimator meta-transformer mapped with their respective meaning. The fact that new columns are added in the beginning of the numpy matrix didn't help (intuitively I was expecting it in the end), but this is ok if I am able to retrieve the names of the given features (assuming I've sent a pandas dataframe as parameter or have given the names in another parameter like "feature_names") and the additional features included by the meta-transformer (maybe with a friendly name too). I will try to fork this and try to work on something to improve this.