[Feature proposal] Allow python scikit-learn wrappers to use callbacks

[mxxun]

Pretty much what it says on the tin. Being able to use xgboost in a scikit pipeline is terribly convenient, and so is being able to finely control learning rate, but at the moment it's one or the other.
As I understand, making wrappers' fit() methods pass callbacks to the train() inside would do the job, but maybe there are reasons not to do it like this, or not to do it at all?

[hcho3]

@mrgutkun Yes, LightGBM does a similar thing, passing a callback to fit() method: https://lightgbm.readthedocs.io/en/latest/Python-API.html#lightgbm.LGBMModel.fit . I'll gladly merge a pull request to enable this.