Implement parameter handling

[ricobruland]

Might consider managing parameters and training configurations in wandb

[dfsbora]

Implemented

• Tracks config info: all the arguments of the train function + architecture (LSTM or transformer)
• Logs training info: loss
• Saves artifacts: dataset, vocabulary mapping and final model

If using data.tex, it will log to "math_latex_project", otherwise it logs to "test_project"

wandb environment is disabled with --disable_wandb on command line

Could be improved:

• flag to save atifacts
• also save info about tokenizer, loss function, and optimizer
• also save evaluation metrics associates to this run