Global Optimization of Low-Thrust Interplanetary Trajectories Using a Machine Learning Surrogate