Watermarking LLMs with Weight Quantization