Unveiling Uncertainty: A Deep Dive into Calibration and Performance of Multimodal Large Language Models