Based on lab flats, generate non-linearity maps for the n1 detector. From the QE acquisitions, we have multiples of 35-read ramps at four wavelengths, some of which go to saturation, some of which don't.

The lab tests showed that most pixels were similar to each other up to ~80% of full well, at which point a significant fraction diverge wildly. If we stayed below that transition we could maybe use per-detector curves, at least to start. Not sure what the ISR stage can deal with.