Towards Practical Preferential Bayesian Optimization with Skew Gaussian Processes