LiNeS: Post-training Layer Scaling Prevents Forgetting and Enhances Model Merging