Torch compile and PyTorch 2.2.0

This project aims to integrate the torch.compile function released in PyTorch 2.0 into DeepChem's PyTorch models, unlocking significant performance improvements for training and inference. By leveraging advanced JIT compilation techniques used in torch.compile, this project will enable DeepChem users to optimize their models for faster execution without requiring code modifications. Initial benchmarks demonstrate promising speedups for various architectures, and further testing will be conducted to ensure broad compatibility and optimal performance gains across DeepChem's diverse model library. This project will ultimately empower researchers and developers with faster and more efficient tools for scientific discovery and innovation.