On the Effects of Quantisation on Model Uncertainty in Bayesian Neural Networks