Update training code, remove legacy libraries, provide basic usage instructions

[ssenan]

This PR revamps the library to make training simpler, more modular, utilizes uv to resolve dependencies, and introduces hydra for config management.

Key changes are as follows:

• Motif analysis has been removed from the training and sampling procedure
• Accelerator has been removed and instead DDP is done using builtin pytorch functionality
• The refactor folder containing earlier efforts to build out the library has been removed
• Training now includes a patience parameter for validation loss improvement along with a minimum training time of
• A number of tests and workflows will likely fail now so these should be revisited in a future PR
• Docs will also need to be reworked inline with theses changes
• This resolves a number of issues (Remove deprecated refactor folder from src package  #291, Update README to replace proposal with usage instructions #292, Generalize Logging function for more flexibility #295, Fix pathing to data files  #298, Use hydra for configs #299, Remove accelerate dependency #300)