Multi-Aspect Cross-modal Quantization for Generative Recommendation