Parallel Bayesian Optimization of Multiple Noisy Objectives with Expected Hypervolume Improvement