Plausibly Problematic Questions in Multiple-Choice Benchmarks for Commonsense Reasoning