Verifying if building native Windows with meson is good enough

Our meson build clearly works for daily dev build, though each time I fix something meson-related, I stumble upon weird rules, or rules not following exactly the autotools-logic or various bugs, etc. So meson is clearly not ready for being considered default and replace autotools fully yet.

I'm actually thinking we should move this decision to GIMP 3.2 and stick to autotools being inconditionally the goto build system for 3.0.

Nevertheless Windows is a special case where the meson build is reputedly much better than autotools because it would be much much faster than the autotools build (on Linux, it is also a bit faster, but it's not as dramatic). Also as a consequence, I think most current Windows devs use meson currently, and probably advising meson as our default build system, at least on Windows, might be one less barrier of entry.

Moreover we finally got rid of one of the worst blocker we had on our meson build a few days ago (the PDB generation). So maybe we can consider this Windows-only policy change.

So the first thing for this MR is verifying the claim of build speed. The gimp-win*-native builds take about 30 min in our runners currently. Let's see how fast is the same with meson.

The second point to verify is that nothing is too broken when building with meson. I already saw that there were directx detection bug, a bug in meson which was never uncovered until now. It's not the best first start.

Let's see how things go.