Use machine.py’s key terms implementation for term extraction

[Enkidu93]

silnlp has native key term handling. For better consistency and maintainability, we should replace this machine.py’s implementation once we’re certain machine.py’s implementation supports everything silnlp supports.

[Enkidu93]

These changes would mostly be at the level of paratext.py. We'd be able to remove most of the code in that class while maintaining the same API. We'd also need to make changes to hugging_face_config.py - namely, removing the adding of localized glosses (which probably isn't a good place to be doing it anyways since there is no artifact in the build folder that lets you see what glosses were used - probably part of the reason we've had a lot of bugs with this) and also potentially using a shared utility function to add the key terms training samples.