Add calculation of derived features for distillable page model
We've trained a simple model for detecting non-mobile distillable pages
based on a set of features derived from a webpage. This change adds the
main part of deriving those features (a bunch of values are calculated
based on some core features of the web page). It does not add the
extraction of those core features from a page.
A test is added that uses json-formatted core features and then
json-formatted output of the training pipelines derived feature
calculation. It then ensures that the calculation here matches that in
the training pipeline.