Let's CONFER: A Dataset for Evaluating Natural Language Inference Models on CONditional InFERence and Presupposition