Force unit in datasets

[wiederm]

I might misunderstand the curation pipeline, but it seems to me that energies and positions are correctly in kJ/mol and nanometer, but the force in some of the datasets is in kJ/mol/angstrom.

@chrisiacovella , can you clarify if I am missing something here?

[chrisiacovella]

This seems to have been a mistake (or an accidental hold over when we actually were using angstroms internally).

I think this is a 2 stage fix:

1. Add unit conversion into the data loader. We have discussed this in the past as an extra measure to ensure all units are consistent, but we had not prioritized it. this will allow us to use existing datasets without our target forces being an order of magnitude smaller.
2. Change units in the datasets. This is easy, but uploading to zenodo is a bit time consuming. As discussed in a separate issue Revamp the datasets/dataset class #267, we can simplify our additional tags in the dataset (that tell us how to parse the numpy arrays) to only being per_system, per_atom, and metadata. changing units, these optional tags, and associated loader, can all be done in the same PR.

I will do a PR for stage 1 now.

[chrisiacovella]

#1 has been address in PR #301. As mentioned in the PR, it would be good to now include a test to ensure proper unit conversion. This will require creating a new test dataset, so I'll do that in a separate PR.