[jvm-packages] the classification model trained by xgboost4j-spark is xgboost in-compatible

[wbo4958]

When saving XGBoostClassificationModel, XGBoostClassificationModel will add the num class into the beginning of the xgboost native model resulting in the model being xgboost in-compatible, which means xgboost native/xgboost python can't load the model trained from XGBoost4j directly.

Although XGBoostClassificationModel provides a way to produce the XGBoost native model by

model.nativeBooster.saveModel(model_path)

But it's really un-convenient to add extra step to get the model.