VQ4ALL: Efficient Neural Network Representation via a Universal Codebook