QAPruner: Quantization-Aware Vision Token Pruning for Multimodal Large Language Models