Pre-fit limit bands, like what we did for median limit

Force to re-minimize Asimov nll during the limit setting. It is found that the nll created for CMS bbyy Asimov dataset is not at the minimum, we should do a further fit the get the correct NLL value