LM Toolkit Refactor

[dcgaines]

Merging toolkit refactor into Banff LM branch for sim testing.

Overview

Replaced all custom models in the language module with language model adapters. Adapters rely on aactextpredict, our new LM toolkit, for the heavy lifting and only need to handle BciPy-specific things like special space and backspace characters and response type properties.

Ticket

Link a pivotal ticket here

Contributions

• Deprecated LanguageModel classes in favor of LanguageModelAdapter classes.
• Consolidated predict methods into the super class, only override when needed (Oracle).
• Renamed KenLM model to NGram to match the aactextpredict package.
• Updated all references to KenLM and LanguageModel classes to match new names/classes

Test

• Ran all pytest cases

Documentation

• Language module README updated. Added links to textpredict repo and AAC adapting arXiv paper.

Changelog

• Is the CHANGELOG.md updated with your detailed changes?
• Not yet.

[tab-cmd]

I'll wait for @lawhead to review since he has more experience here. I would keep the base class as LanguageModel or BciPyLanguageModel, but the others could be annotated as Adapters extending from that. We may want our own Uniform here without an adapter. I understand why you need it in the toolkit, but it's simple enough to keep here, and it could be a good example of how to build an LM in BciPy.

The toolkit doesn't seem to work for 3.10.6. >=3.7,<3.11?

Also, some linting errors!