Fairness for Cooperative Multi-Agent Learning with Equivariant Policies