Input_training(cross_validation = TRUE) with current xgboost
releases by replacing the caret::train(method = "xgbTree") tuning path with
native xgboost::xgb.cv() while preserving the MethScopeModel return
structure used by PredictCellType().