WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation