Temporal Aware Pruning for Efficient Diffusion-based Video Generation