[Tokenizers] Implement WordPiece algorithm

[ericstj]

The WordPiece algorithm should be added to Microsoft.ML.Tokenizers. WordPiece algorithm is the basis for BERTTokenizer-based models. Needed for E5

We can see reference implementations in
https://github.com/microsoft/BlingFire (MIT license)
https://github.com/huggingface/tokenizers (Apache license)

The paper which it's based on:
https://arxiv.org/abs/1609.08144
https://arxiv.org/pdf/1609.08144.pdf