Comprehensive reference on input files

[FranklinHu1]

There are a lot of settings associated with setting up an experiment, some of them being general purpose ML parameters (e.g. learning rate, batch size, epochs, etc.) and some that are more bespoke to DFTBML (e.g. number of spline knots, spline cutoffs), and it is important for users to have access to this information. Additionally, the configuration .json files contain many unused arguments that were left in during the development of DFTBML but should be removed. To that end:

• Remove dead/unused arguments in configuration files
• Test to ensure experimental results are replicable
• Compile a reference document on what all of the parameters do and best practices for setting them