Switch to base toolkit package

[mattwthompson]

I'm not necessarily sure this should be merged; the objective here is to see what happens in tests if NAGL is provided openff-toolkit-base and if anything is missing and needs to be packaged.

Changes made in this Pull Request:

• Switch to using openff-toolkit-base

PR Checklist

• Switch to using openff-toolkit-base
• Add back any dependencies not brought in by openff-toolkit-base
• Tests?
• Docs?
• CHANGELOG updated?
• Issue raised/referenced?

[mattwthompson]

Surprisingly only people like me who don't think things through, the tests here fail because AmberTools is needed for AM1-BCC charges when openeye=false. I'd like to be able to hand-wave this away and argue it should remain a test-only dependency. That doesn't seem satisfactory, though, since code like

MoleculeRecord.from_openff(
            offmol,
            generate_conformers=True,
            partial_charge_methods=["am1bcc"],
            bond_order_methods=["am1"],
        )

bubbles up with the expected ValueError: No registered toolkits can provide the capability "assign_partial_charges" .... Is this viewed as an acceptable outcome? I mean, if the user doesn't have something installed that provides AM1-BCC, I don't know how they'd expect a method named MoleculeRecord.from_openff to work. I don't know the code well enough to asses if MoleculeRecord is core functionality or not. Another option here would be to catch and handle this case.

[lilyminium]

MoleculeRecordwas core functionality, but will stop being so in 0.3.x (everything in storage is going away) and the replacement will hopefully be clearer in that you need to have a charge method available -- the replacement will be something like openff.nagl.training._loss.ChargeReadout.compute_reference. That being said, I think you could do MoleculeRecord.from_openff(partial_charge_methods=[]) currently and it would still work -- so I'm happy for the user to try doing something am1 related and bump into this error if they don't have requisite toolkitwrappers enabled.

Would it be alright if I put this temporarily on hold until I push more into https://github.com/openforcefield/openff-nagl/tree/sqlite-to-pyarrow?

[mattwthompson]

Happy to pick this up whenever those changes are pushed through

[lilyminium]

Closing as superseded by #45.