Benchmarking Constraint Inference in Inverse Reinforcement Learning