Efficient Fine-Tuning and Concept Suppression for Pruned Diffusion Models