Towards Privacy-Aware Bayesian Networks: A Credal Approach