Migrate repo to modern machine learning benchmarking standards

[jerrylin96]

Specifically, making use of PyTorch Lightning, Wandb, and fixed seeds for reproducing results (some changes may be unrealistic due to constraints from already written paper)