QDepth-VLA: Quantized Depth Prediction as Auxiliary Supervision for Vision-Language-Action Models