Enhance fast_predict by requiring fewer arguments (just the model)

[mulhod]

It would be nice if we could make feature_info, trim_min, trim_max, and trim_tolerance attributes of the Modeler object. We can still expose these things in the fast_predict function. We could use the values if they are provided and not None. Otherwise, we check for the presence of these attributes and, if specified, use them instead. For feature_info, it must be able to be found as an attribute if None/not specified in the call to fast_predict. For trim_min/trim_max, if it's both None/unspecified and cannot be found as an attribute, the default value None will be used for (no trimming). For trim_tolerance, if it's both None/unspecified and cannot be found as an attribute, the default value of 0.4998 will be used.

This should make it possible to call fast_predict with only the data and a Modeler object.

[mulhod]

Actually, I will look into adding train_predictions_mean, etc., as well. Anything, basically, that could be added as an attribute.