Implement optimal 2D background subtraction

Algorithms exist to optimally subtract the sky background from 2D spectra that account for optical distortions and other effects. These effects create non-linear relations between spatial/wavelength coordinates and pixel-based detector coordinates. Careful mapping of these effects allow one to build a sky background model with sub-pixel resolution. One such method is described in [Kelson (2003)](https://www.jstor.org/stable/10.1086/375502). That algorithm should be implemented in `specreduce` and possible updates explored.

Start by reading the specreduce codebase and the Kelson (2003) paper linked in the issue to identify the relevant 2D spectrum reduction entry points. Define the scope of the algorithm and any updates to explore, then verify that the implementation models sky background with sub-pixel resolution while accounting for the described coordinate distortions.

Written by the indexing model from the issue text.