Hyperopt search should allow optimizing on any metric, not just r2_score or roc_auc

[mcloughlin2]

The loss function used in a hyperopt search is always based on R-squared for regression models and ROC AUC for classification. Since we sometimes want to optimize for a different metric (e.g., balanced accuracy or MCC for an imbalanced classification dataset), we should allow users to specify the metric used by hyperopt using the model_choice_score_type parameter.