Storing various trained ML models.

[nquetschlich]

The API could be improved so that the constructor of the ml.Predictor takes a trained model (or a path to one) as a constructor argument. That argument may default to the pre-trained model that we also distribute, but it may was well be set to something a user has trained themselves.

Originally posted by @burgholzer in #130 (comment)

[nquetschlich]

Closed since we do not have concrete plans to implement this.