Compute generalized advantage estimates by recursively combining rewards, value predictions, discounts, and GAE decay.
