HyperQuant: A Rate-Distortion-Optimal Quantization Pipeline for Large Language and Diffusion Models