Forecasting with gradient boosted trees: augmentation, tuning, and cross-validation strategies