train.py enhanced checkpoint resuming

[theobjectivedad]

• Adds automatic checkpoint resuming logic to train.py
• Updated train progress bar to include epoch and train_loss
• Added an __init__.py to make it easier to invoke gpt4all code from other Python wrappers; there is a larger discussion that should be had regarding whether python files should be moved into a subfolder and make the project a proper Python package.
• Configuration files should be backwards compatible

To start a new training session use the following configuration:

checkpoint: ~

As before, to force a specific checkpoint:

checkpoint: "foo/gpt4all-7b-hf-lora/step_135000"
train_args:
  resume_from_checkpoint: "step_135000"