Fixed bug that meant range_cnr images were written instead of std_cnr images

There was a bug that meant that the "range-cnr" images were written out instead of cnr images defined as std(signal)/std(residuals), where std(signal) is estimated from the predictions. As far as I can tell this bug must have been present for a long time. It is not a serious bug in that it only concerns a "diagnostic" output metric, and all relative CNRs will have been the same. But the scale has been wrong, which means that CNR values should not be compared across the bug fix.