Compact Neural Volumetric Video Representations with Dynamic Codebooks