This will calculate the rescaling needed when covariates to be included in
the scale of the detection function are "too big". Based on code from
optimx.
rescale_pars(initialvalues, ddfobj)starting values for the optimisation
detection function object
David L Miller
Derivative-free methods like nlminb are sensitive to the parameters being poorly scaled. This can also cause problems for quasi-Newton methods too (at least, bad scaling won't _help_ the optimisation). So here we rescale the parameters if necessary (unless we already got scaling from control)