Add sklearn.preprocessing.LabelEncoder

[yujonglee]

We need to make it easy to use sklearn.preprocessing.LabelEncoder, or implement it for pipeline interface.

Need some good warning if labels is not encoded. See test_model_grade.py

[yujonglee]

This kind of thing can be useful in human eval too, since typing full label name is cumbersome.

[yujonglee]

Maybe we don't need to use this, but adapt its interface.

get instance - fit - transform - inverse_transform makes sense.