Assumption-light and computationally cheap inference on inequality measures by sample splitting : the Student t approach