Atropisomer CX Smiles generation is not canonical

[ricrogz]

I hit this while working on the failed test in #7426:

from rdkit import Chem

atropTestFile = '../Code/GraphMol/FileParsers/test_data/atropisomers/RP-6306_atrop1.sdf'
mol = Chem.MolFromMolFile(atropTestFile)

# Strip conformers so that we can't accicentally use them
mol.RemoveAllConformers()

# This is the atrop bond
bond = mol.GetBondWithIdx(3)
assert bond.GetStereo() ==  Chem.BondStereo.STEREOATROPCW

ps = Chem.SmilesWriteParams()

cxsmi1 = Chem.MolToCXSmiles(mol, ps, Chem.CXSmilesFields.CX_ALL_BUT_COORDS)
print(f'{cxsmi1=}')  # cxsmi1='Cc1cc2c(C(N)=O)c(N)n(-c3c(C)ccc(O)c3C)c2nc1C |wD:11.11|'

mol2 = Chem.MolFromSmiles(cxsmi1)
cxsmi2 = Chem.MolToCXSmiles(mol2, ps, Chem.CXSmilesFields.CX_ALL_BUT_COORDS)
print(f'{cxsmi2=}')  # cxsmi2='Cc1cc2c(C(N)=O)c(N)n(-c3c(C)ccc(O)c3C)c2nc1C |wU:10.9|'

# Smiles canonicalization should be stable
assert cxsmi1 == cxsmi2, 'canonicalization is not stable!'  # AssertionError: canonicalization is not stable!

[greglandrum]

Yeah, we're going to have to take either the canonical atom ranks or the atom/bond output orders into account when selecting which bonds to wedge.