Update tokenizer to accept new characters for Huggingface models

[johnml1135]

Related to the effort here: sillsdev/silnlp#171

[johnml1135]

This is to add new characters to NLLB 200 to enable both random one-off characters and entirely new scripts. Current status:

• This is working in SILNLP here: Update tokenizer with missing characters silnlp#171
• Tokenization can either add individual characters or use Sentencepiece to tokenize based upon frequency of tokens in words in the corpora

Things to do:

• Check if freezing the model and adding another layer helps with Blue score, especially with new scripts as in this threadthub thread
• Implement in Machine.py with some testing
• Add basic controls through YAML files for enabling, size of vocab, etc.