On the Policy Iteration Algorithm for Nondegenerate Controlled Diffusions Under the Ergodic Criterion