Interpolation Model impossibly slow

[okennedy]

View definitions give Spark a headache. These should be getting universally compiled out by Mimir before the query gets handed over to spark. It looks like the Interpolation model's training process doesn't do so.

See e93e2e6 for a discussion of how this problem was solved for the Classifier model.

[okennedy]

@mrb24 has worked on this and improved the situation