Support quantile regression via compute_model_predictions

[dholstius]

Adds support for quantile regression, though probably not in the most efficient way. There's only one line in pred_grid.rq that differs from pred_grid.lm, but I'm not sure how best to abstract the similarities.

Started this because I'd like to see support for time-series "filters", and would be glad to try implementing after I understand the thinking behind compute_model_prediction a bit more. Then I might be able to help implement something like local quantile regression ... ?

Would one-sided time series filters (running means, medians, etc.) best be implemented as "pseudo-models"? Or are they their own beasts?