G2p-id-py

Breaking changes:

- Add glottal stop in between consecutive vowels
- Collapse all k graphemes to k phoneme

- Wrapped ONNX Inference Session to allow pickling, useful for model serialization with tokenizers and/or parallel processes where serializability is needed.

- Fixes https://github.com/bookbot-kids/g2p_id/issues/16.
- Pinned NLTK version due to backward incompatibility with >=3.8.1

Code refactors as suggested by Pylint.

Fix ONNX Runtime-related issues.

- Swapped `word_tokenize` to `TweetTokenizer`.