ci[windows]: revert to windows 2019 builder to try to fix Xerces C

[dancergraham]

PR to trigger build and test the solution

[asmaloney]

Thanks Graham. That didn't seem to do the trick, but it was something I hadn't tried yet!

[dancergraham]

agh oh well it was worth a try. I will try to compare my build scripts with yours at some point to look for a fix. I feel out of my depth with make files, C++, etc but I can always hack and hope.

[asmaloney]

I appreciate the attempt! It is a simple(ish) script aside from the conda/Windows setup - I tried to keep all the OS-specific stuff isolated so that the configuration didn't need to be special-cased & I didn't have to add more options to the matrix, but maybe that's not possible anymore (for some unknown reason).

As I mentioned in the issue - I don't even know where the problem is - could be the CI runner, miniconda, CMake, or something else entirely. I do know it was working one day, and without changes, it started failing which is really frustrating.

I'm not a Windows person - I only use it via VM when I absolutely have to - so I don't have a great mental model of where things are on the system or what's going on. I've only ever used conda (or any C++ "package manager") via CI.