Efficient Pruning of Text-to-Image Models: Insights from Pruning Stable Diffusion