V2.0

1. New configuration file format with more options on inputs and event topology.
2. Allow for several different inputs, including global inputs for additional context.
3. New Regression and Classification output heads for performing per-event or per-particle predictions.
4. Gated transformers and linear layers for more robust networks. Less hyperparameter optimization.