Large Language Models as an Indirect Reasoner: Contrapositive and Contradiction for Automated Reasoning