Pareto Active Learning with Gaussian Processes and Adaptive Discretization