Rethink generation of 2D wavelength, slit-position, order maps from full dispersion solution

[thespacedoctor]

This task is by far the most CPU-intensive and slowest section of the pipeline reduction cascade. 2 ideas to speed up the (currently brute force iteration) the code:

1. Fit 2 bspline surfaces (wavelength and slit-position) seeded with a grid of wavelength and slit-position. This could allow interpolating of wavelength/slit-position to the centre of each detector pixel.
2. When fitting the full dispersion solution (x and y as functions of wavelength, slit-position and order) also then fit the opposite (wavelength and slit-position as functions of x and y and order).

Solution 1 might be a quick win, so trying this first.

[thespacedoctor]

bspline surfaces

• [x] seed a wavelength surface and plot result
• [x] seed a slit-position surface and plot result

[thespacedoctor]

In the end I switched from a bspline surface fit to nearest-neighbour (cubic spline) interpolation seeded with a grid of wavelength and slit-position.

The 2D wavelength/slit-position/order image maps are now generated 6-8 times faster (10-12 mins down to ~90sec) and are 100 times more accurate.