Can Language Models Take A Hint? Prompting for Controllable Contextualized Commonsense Inference