ADR reference wavelength should be grating dependant

[fpavogt]

Currently, it seems that the ADR aligns all the cube wavelength slices with respect to 5000 Å (see wifes_adr.py, l.84).

This sort of makes sense for e.g. B3000+R7000 observations, but not for e.g. the U7000 grating. It would be much better to have a reference wavelength dependent on the grating used (i.e. in the middle of the transmission window) to maximize the number of spaxels with data.

In fact, the best way to do this would be to bring this keyword back to the surface (in reduce_blue/red_data.py) to give users the choice of a) perfectly consistent red & blue cubes with less spaxels, or inconsistent red&blue cubes each maximzing the number of spaxels.