Enhancing Language Multi-Agent Learning with Multi-Agent Credit Re-Assignment for Interactive Environment Generalization