Add support for deep learning models into Mudlark

[Michael-Stewart-Webdev]

At the moment Mudlark is fully heuristic based, with simple_normalisation being the only normalisation method. Another method could be added (something like deep_learning_normalisation that has support for Huggingface models. The Mudlark interface (e.g. the code in main.py) would need to be adjusted to have an option allowing the user to specify what type of normalisation model they want to use, and if they're using deep learning, then the user could also specify a URL of a Huggingface model or something along those lines.

I believe Tyler (@4theKnowledge) has a lexnorm model on Huggingface already (or at least is planning to put one on there).