Returns a learner that fits OLS on the control units. With many covariates relative to the number of controls, OLS overfits — which is exactly the case that makes the fit-once / fixed-residual shortcut anti-conservative and that refit-per-permutation handles correctly. Useful for demonstrating that contrast; ridge is the better default in practice.