Let's take this opportunity (lots of new features in revscoring that various models might use) to update the libraries we rely on and build up support for new versions of python.
New features:
- Better handling of pt dictionaries (pt models)
- Portuguese words to watch (pt models)
- Improved tokenization (all models)
Extend support for python 3.7 and 3.8.
Use the lastest version of sklearn, numpy, and scipy.