An assumption‐free convergence analysis for a perturbation of the scaling algorithm for linear programs, with application to the L 1 estimation problem