PyBel based method get_smiles, results in inconsistent outputs

[CalCraven]

Bug summary

What were you trying to do and what happened instead? Please copy and paste the stack output

cpd = mb.load("CC(=O)Oc1ccccc1C(=O)O", smiles=True)
cpd.get_smiles()

This calls pybel and some of the infer bond_order stuff, and one of our tests is catching a flaky behavior with aromatic rings, which is not handling implicit hydrogens. The test right now is written to account for that behavior, which seems to occur on different versions of Python for our Ubuntu tests. This will need to get looked into more deeply and potentially lead to removing this method in favor of the to_smiles() method from RDKit.