On a heuristic stopping rule for the regularization of inverse problems by the augmented Lagrangian method