Estimate policy state values from complete sampled returns without requiring an environment transition model.
