Using Bayesian deep learning approaches for uncertainty-aware building energy surrogate models