Towards Certification of Uncertainty Calibration under Adversarial Attacks