Stanza Tokenizer

[koaning]

This is the first implementation of the stanza tokenizer and adresses #23. There's some cool things about this implementation:

0. We add a tokenizer. Hopefully this works better for some languages than our basic WhitespaceTokenizer.
1. We automatically add a lemma to each token so that means that the CountvectorFeaturizer inside of Rasa can pick them up.
2. We automatically add a pos to the data property of each token. I'm using the same technique as the SpacyTokenizer so that means that you should also be able to use them from the LexicalSyntacticFeaturizer.

I still need to figure out a nice way to test this on CI because the model that I'm downloading is fairly big. I also need to figure out if the caching mechanisms are appropriate but it is ready for a first review. There's a few things that need to be checked before merging.

• Can we make this dependency optional? It also install a lot of pytorch and that might not be relevant to everybody.
• Can we confirm the caching dir works as we expect?
• Can we confirm/add a test that ensures that the countvectorizer/lexicalfeaturizer behave differently with stanza around.

[koaning]

Edit, since we install via pip+git it's going to be somewhat tricky/odd to make the stanza dependency option. Work for later.

[koaning]

The docs still need to be appended. It's annoying but for Windows you need another pip install command to deal with the pytorch dependencies.

[tabergma]

Looks great 💯

[tabergma]

Looks great 🚀