Provable Policy Gradient for Robust Average-Reward MDPs Beyond Rectangularity | Read Paper on Bytez