TAPDA: Text Adversarial Purification as Defense Against Adversarial Prompt Attack for Large Language Models