GLoRE: Evaluating Logical Reasoning of Large Language Models