Integrate binding model into stacked_tokenizer

[lgessler]

• Move most binding code into lib/
• Move some data files into data/
• Integrate new ML-based binding into the detok section of the stacked tokenizer
• Comment eval_binding out of eval.py (can fix this if we need to, we just changed the signature of run_eval because of the need for both gold and orig for a train set or test set)

To test the PR:

cd lib
# train the model
python binder.py xgboost --train_list=onno+ephraim+victor+cyrus
python stacked_tokenizer.py ../eval/plain/aug_bind_uddev.txt -d