Change the way we calculate fiberNorms (not the same as PIPE2D-1442)

The current constructFiberNorms.py code is built on the usual SpectralCalibTask framework, but I think this is inappropriate in this case as all the information needed (or desired) is in the pfsArm files. What we want to know is the normalisation that needs to be applied to those spectra to normalise them to the JEG lamp or sky; I think with Paul's current approach that the JEG lamp is probably the way to go, but that's more PIPE2D-1442 territory.

A few other points:

• The current code reprocesses from raws (or postIsrCcd images, if you set rerunISR=False rather than usePostIsrCcd=True, which seems different from the way that this is handled elsewhere). This is awkward if you want to ask questions about experimental ways of processing the frames (e.g. if you are working on scattered light).
• We need access to fiberNorms for exposures that may never be ingested, to monitor the throughputs, as they provide important QA
• the fiberNorms need to be for the entire instrument, not just for one spectrograph. Currently it's probably OK to do this arm-by-arm, but the QA will need to look across the arms. We should probably include the ENGINEERING fibres, as in the longer-term we will probably want to monitor them too, but I forget how we handle that in pfsArm files; I'd do the same thing here.

So I think we need a rewrite to process the pfsArm files for a set of visits to extract the needed information (we'll need to e.g. median filter the .flux data, but that's small enough that the usual code designed to handle many 16Mpixel images isn't needed), then write the results to a set of butler-visible files that can later be ingested.