Reducing Storage of Pretrained Neural Networks by Rate-Constrained Quantization and Entropy Coding