Task Vector Quantization for Memory-Efficient Model Merging