Merge CalBERT-rewrite with Main

[aditeyabaral]

CalBERT-rewrite

This set of changes includes a complete rewrite of CalBERT and includes classes for the model, dataset and the Siamese Pretrainer.

Changelog

• Complete rewrite in the most popular deep learning framework (IMO), Pytorch allows CalBERT to be used more extensively and easily. Support added for CalBERT to be used as-is, or imported as a layer in your own models.
• Separate Torch-based class for handling CalBERT dataset creation with negative sampling included as well
• Siamese Pretrainer is now a new class by itself, allowing you to pre-train any Transformer using CalBERT's Siamese Pre-training. Supports checkpointing, different contrastive losses (some not in the paper!), saving and loading of models and more.
• A simple train.py script to help beginners get started with training CalBERT
• CalBERT can now be installed via pip