Change input data structure from named tuple to a class with slots

[wiederm]

Pull Request Summary

This PR addresses the issue #286.

We had two data structures (NNPInput and NNPInputTuple) to pass information from the dataset to the models. The NNPInput class was designed as dataclass and did input validation and casting to dtype/device during training. The dataclass then constructed a NamedTuple that contained all fields of the dataclass, which was then passed as input to the model --- in other words, the model only ever got the NamedTuple as input. To allow working with Jax models, the Named Tuple contained either jax arrays or pytorch tensors depending on passed keywords.

NamedTuples are useful during training, but in MD simulations the immutable datastructure leads to overhead.

Key changes

• Change NamedTuple datastructure to a Python class with slots, and remove the double structure dataclass/namedtuple that was used for convenience.
• change names in NNPInput and Metadata classes from E and F`` to per_system_energyandper_atom_force`

This PR proposes the following changes:

Pull Request Checklist

• Issue(s) raised/addressed and linked
• Includes appropriate unit test(s)
• Appropriate docstring(s) added/updated
• Appropriate .rst doc file(s) added/updated
• PR is ready for review

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 84.46602% with 32 lines in your changes missing coverage. Please review.

Project coverage is 84.97%. Comparing base (442df66) to head (a6a791c).
Report is 26 commits behind head on main.

Additional details and impacted files

[chrisiacovella]

I think this is a good change; one concern, you've now started to introduce "per_system" instead of "per_molecule", but it's not consistent. I think that should be changed over fully before merging, otherwise that is going to be very confusing.