Support for alternative metrics for `oob_dropcol_importances`

[ValterH]

The default implementation of oob_dropcol_importances sets the oob_score attribute to true. This results the method always returning the default ($R^2$) score. Adding a keyword argument for metric of interest would greatly improve the method's flexibility. I would be happy to submit a PR as the implementation would be quite simple.

Desired outcome: drop_imp = oob_dropcol_importances(model, X_boot, y_boot, oob_metric=negative_rmse)