Atom stereochemistry setter does no input validation

[mattwthompson]

Describe the bug

I think one would have to try to run into this problem, but at very least the docstring is incorrect

https://github.com/openforcefield/openff-toolkit/blame/e71757033cd6a81a1e188d62e8b5d80186c61934/openff/toolkit/topology/molecule.py#L446-L460

To Reproduce

>>> from openff.toolkit import Molecule
>>> water = Molecule.from_smiles("O")
>>> water.atom(2).stereochemistry
>>> water.atom(2).stereochemistry = "The Mythical California Zephyr"
>>> water.atom(2).stereochemistry
'The Mythical California Zephyr'

[j-wags]

Agree, if the docstring says this can only have certain values, then we should make sure that's in line with the behavior

[j-wags]

I made a relevant comment in #2086:

I'm weakly against validation [in molecule.py], since we don't know which custom cheminformatics toolkitwrapper users might have, and there are valid stereo definitions beyond R and S. If we really want to validate, we could have the cheminformatics wrappers raise an exception if a stereochemistry they can't handle comes through.

[Yoshanuikabundi]

I replied to that comment:

I'm weakly against validation here, since we don't know which custom cheminformatics toolkitwrapper users might have, and there are valid stereo definitions beyond R and S. If we really want to validate, we could have the cheminformatics wrappers raise an exception if a stereochemistry they can't handle comes through.

IMO we should define everything that's valid and let the cheminformatics wrappers translate it into their own language - even if that makes the wrappers complicated. This is important so that Molecule objects created with one toolkit can be processed with other toolkits. Users implementing custom toolkits benefit from knowing the API they're targeting.