These updates have had four beneficial effects which have all been tested on a real DECam data set:

1. Improved run-time on spatially-varying calculations due to decrease in number of grid elements used
2. Improved robustness of Zogy due to removal of division-by-small-number issues and enforcing same-centroid-pixels for both PSFs
3. Improved robustness of AL Decorrelation due to same as (2)
4. Enabling "doPreConvolve" with decorrelation and Zogy from the imageDifference.py config.

Passed Jenkins: https://ci.lsst.codes/job/stack-os-matrix/26729/

Hi Russell, as you suggested earlier today I moved all of my algorithm bug-fix code to a different branch (out of the imageDifference refactor) and now I'm hoping you'd be willing to review it? It is not a very large amount of code. Thank you!

Overall this looks nice. I had a few small suggestions.

Thanks very much for the review, Russell Owen. I have made the changes you suggested and/or responded to your comments/questions.

Re-ran jenkins just to double-check, and it pass the ip_diffim build:

https://ci.lsst.codes/job/stack-os-matrix/26737/

so I will now merge.