OpenFF Toolkit 0.12.0 breaks some tests

[mattwthompson]

=========================== short test summary info ============================
FAILED openff/qcsubmit/tests/test_datasets.py::test_dataset_linear_dihedral_validator - ValueError: atomic number must be positive, given 0.
FAILED openff/qcsubmit/tests/test_datasets.py::test_torsiondrive_unqiue_settings - ValueError: atomic number must be positive, given 0.
FAILED openff/qcsubmit/tests/test_factories.py::test_torsiondrive_torsion_string - ValueError: atomic number must be positive, given 0.
FAILED openff/qcsubmit/tests/test_workflow_components.py::test_wbo_fragmentation_apply - assert 0 == 4
 +  where 0 = ComponentResult(name=WBOFragmenter, molecules=0, filtered=1).n_molecules
FAILED openff/qcsubmit/tests/test_workflow_components.py::test_pfizer_fragmentation_apply - assert 0 == 4
 +  where 0 = ComponentResult(name=PfizerFragmenter, molecules=0, filtered=1).n_molecules
====== 5 failed, 470 passed, 5 skipped, 30 warnings in 1718.32s (0:28:38) ======

The relevant change is at least https://github.com/openforcefield/openff-toolkit/pull/1480

[mattwthompson]

At first I thought this might not warrant a new release, but now understanding the change I think it does. Anything touching that torsion-building logic might now fail.

What I have in mind is

• Re-build the current release (0.4.0) to openff-toolkit-base =0.11 or something similar
• Merge #210
• Make a new release (0.4.1) pinned to openff-toolkit-base =0.12

I can handle executing all of this but would like feedback prior

[j-wags]

This plan sounds perfect. Thanks for the quick action on this, @mattwthompson!