Adding Resume Training Functionality

[yousefelsharkawy]

• Saved the data loader and optimizer states to resume training from where the model stopped
• There are functions to save and load the rng states, but they seem to work on CPU only (refer to https://pytorch.org/docs/stable/random.html#torch.random.fork_rng). HOWEVER, saving and loading the model, optimizer, and data loader states seems to be sufficient to capture the state of training (I Trained the code for N steps on a single run, then did that again but for multiple runs using the resume training functionality, and I got the same loss and norms)