Inconsistent application of Astier's amatrix in brighter fatter correction

I'll push this change through the review process, and I'll try to work through the math today and tomorrow to see if this fix is hiding other mistakes.

Here is a plot of the correction with the negative sign change, showing that it is correcting in the "right direction"

I can confirm that Chris's fix works in flipping the A matrix and results in a proper correction. But while it does work, I don’t know if it is just a fix for an issue elsewhere

@Alex Broughton will try it.

If it's just a simple minus sign, then it probably just needs to be added here: https://github.com/lsst/cp_pipe/blob/main/python/lsst/cp/pipe/makeBrighterFatterKernel.py#L304 

As that's a trivial change, I've made it on a ticket branch here.  If you can test it out to confirm that it fixes the issue, I'll make a PR and get it reviewed etc.

I thought that I had generated a similar-looking kernel with both the mean cross-correlation and A-matrix paths, but you're doing a much more in-depth study that I did.