Add the machine learning model from "Automatically Selecting Striking Images for Social Cards"

[shawnmjones]

Now MementoEmbed still applies the same scoring equation it always has. In Automatically Selecting Striking Images for Social Cards, we found that Random Forest and base image features worked best for predicting the same image chosen by the author of a web page.

We should incorporate that model before I forget how this all works.