Add a function to estimate covariance matrix of output image

It would be nice to add a function that estimates a covariance matrix of output pixels for any of the reprojection functions. We could make this a 'brute force' way which uses Monte-Carlo simulations to reproject a number of noise images, and empirically measure the covariance between output pixels. This would not be the most efficient way to do it but could work for *all* of the reprojection algorithms.

Start by inspecting the reprojection functions and how they produce output pixels. Define the function's interface and the Monte Carlo noise simulation and covariance behavior before implementing it. Done means covariance estimates are available for every reprojection algorithm.

Written by the indexing model from the issue text.