Implementing dual-convolution. This required reworking those
functions involved with calculating and solving the equations; moved
these into a separate file and made various other cleanups (e.g.,
assertions for pmSubtractionKernels). Removed the normalisation
(central pixel) term from all kernels, because this shouldn't be in
the second kernel (there's only one normalisation term between the two
kernels); the normalisation is treated explicitly in the equations,
along with the background (still only a constant background is
supported for now, but there's a lot of support for a polynomial
background for when I get around to putting it in the equations).
Tested the single convolution, and it's working; same results as
before, apparently. Haven't tested dual convolution yet.
