pymatgen structure creation patch

[laserkelvin]

This PR resolves #304 by refactoring PeriodicPropertiesTransform in two ways:

1. We will reuse a serialized structure if it is interpreted as a pymatgen.core.Structure for the periodic boundary neighbors calculation, and return directly.
2. If the above condition fails, we'll continue down the old path, but Lattice.from_parameters is now passed vesta=True in all cases if creating from lattice parameters. This set up to happen in both the transform, as well as the subroutine matsciml.datasets.utils.make_pymatgen_periodic_structure.

The impact of this change is likely only with datasets that do not have a cell attribute in their data. e.g. LiPSDataset and Open Catalyst expose cell which is used for the Lattice creation, and if the fully qualified lattice matrix is given then the resulting lattice is consistent. Only when no cell is available is when Lattice.from_parameters is called.

A related issue is #267, if and when we get to it. There would not be inconsistencies if the pymatgen.core.Structure is created from first principles, as opposed to loading a serialized one.

[melo-gonzo]

Thanks for the updates. Glad this is all sorted.