INFELM: In-depth Fairness Evaluation of Large Text-To-Image Models