Update RDKit versions used in tests

[mattwthompson]

Work around

• Pointer freeing error when using rdkit.Chem.Draw in 2024.03.6 and 2024.09.1 conda-forge packages for ARM Macs rdkit/rdkit#7890
• Stub file(s) contains null bytes rdkit/rdkit#7221 (and a few other stub-related issues)
• Some changes somewhere in 2023.3

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.73%. Comparing base (df251d3) to head (bad5456).
Report is 1 commits behind head on main.

Additional details and impacted files

[j-wags]

Looks like this is getting cheminformatics-y. Mind if I switch this on to my plate to dig into the changes?

[mattwthompson]

Please feel free to! I think it might come down to why the behavior around #1942 (comment) changed in some of the matrix but not the rest. I'm happy to wash my hands of this if you think you can read the tea leaves

[j-wags]

Cool, I'll take this over then!

[j-wags]

The molecules that are newly failing (er, passing) stereo tests are somewhat exotic rings with nitrogens. Loading both with RDKit 2022 leads to undefined stereo errors on nitrogens[1], but not any more in RDKit 2024. Without fully delving into the context of all our past decisions about interop, raising fewer stereo errors here is a GOOD thing, since the molecules are 3D to begin with, so the idea that they're missing stereo is nonsensical. So I have no hesitation removing molecules from this xfail list.

[1]

openff.toolkit.utils.exceptions.UndefinedStereochemistryError: Unable to make OFFMol from RDMol: RDMol has unspecified stereochemistry. RDMol name: DrugBank_7124Undefined chiral centers are:
 - Atom N (index 43)

[j-wags]

I'll consider my final round of cleanups to be equivalent to my review and will merge this PR once I've made them.

[mattwthompson]

Great - removing skipped molecules meshes with my surface-level understanding of those "failures" being improvements