KekulizationException in tautomer canonicalization

[d-b-w]

Describe the bug
There is a KekulizationException when canonicalizing tautomers for some types of molecules. This seems to be new since the sprint release. I've seen it in a couple of places, but the commonality seems to be with non-aromatic rings with two nitrogens and one external methyl group.

Here are a couple of examples:

To Reproduce

from rdkit import Chem 
from rdkit.Chem.MolStandardize import rdMolStandardize  

for smi in ['NC1=NC=NC(C)=C1',
            'CC1N=CN(C)C(=O)C=1',
            'CC1=CC=CC(=O)N1C']:
    mol = Chem.MolFromSmiles(smi)
    rdMolStandardize.TautomerEnumerator().Canonicalize(mol)

Expected behavior

some sort of answer :)

Screenshots
If applicable, add screenshots to help explain your problem.

Configuration (please complete the following information):

• RDKit version: 2022.09 release. Also tested with a build at adfdeca (head of origin/master on November 20)
• OS: Darwin, Linux and Windows
• Using an internal Schrödinger build, also built locally

[d-b-w]

Looks caused by #5402 - if I revert that commit these structures give answers

[greglandrum]

Yeah, there's a problem with that change. I'm looking into it, but it's non-trivial (as one would expect with anything connected to tautomerism)

[d-b-w]

Would it make sense to catch kekulization errors and continue at this point? I think it would be best if all transformations were guaranteed to always produce valid structures, but that seems like a hard guarantee to make!