In order to train a learning-to-rank model for mediasearch, we need to gather the scores for different search signals

The way to do this is to create a featureset in elasticsearch that calculates scores for different search signals in the same way that current mediasearch does (we can create future versions based on different versions of mediasearch, but let's begin with what we have now)

See https://elasticsearch-learning-to-rank.readthedocs.io/en/latest/building-features.html