Understanding (Un)Intended Memorization in Text-to-Image Generative Models