CommVQA: Situating Visual Question Answering in Communicative Contexts