Hybrid CNN-ViT Framework for Motion-Blurred Scene Text Restoration