Inverse Constraint Learning and Generalization
by Transferable Reward Decomposition