Extend CI with Windows (MSVC x86, MSVC amd64, GCC i686, GCC x86_64 and Clang)

[friendlyanon]

Seeing how you added CI testing, would you also be interested in adding Windows to the tested platforms?

I mentioned in #1 that I have a fork with a test binary produced using the GCC builtins. This would need to be committed into this repo as well and the testing code needs to use that as a reference, since MSVC has no builtins to check against.

The fork can be found here https://github.com/friendlyanon/jtckdint and I have been tweaking it a lot, mainly to experiment with things, so please let me know if you like the look of things and I will create a PR with only the necessary changes.

Should also be mentioned that for GCC i686 and x86_64 I am using my minimal fork of Chris Wellons' w64devkit. I build the GCC and MinGW libraries in GitHub Actions. These can also easily be used in CI. Please let me know if you agree.

[friendlyanon]

So, I'm trying to work on upstreaming the related changes from my fork, but I'm having trouble.

I have created a C89-ish file that reproduces the test.bin data for testing implementations without GCC builtins: https://github.com/friendlyanon/jtckdint/blob/master/corpus.c
The output of this tests perfectly with GCC builtins (duh!) and C(++)11 implementations of my fork, but not with (practically) unaltered upstream header: https://github.com/friendlyanon/jtckdint/commit/28b960cecb6691efe34f6f75bfa3fbc123077087#diff-b8d595a699e7694d818d745fd1d7056bb853a61a15479d8bbd6bc90faf1bf5c2

Mismatch @ 0x7FCA
  Actual:   (1) 1
  Expected: (0) 1
  Types: T = uint8_t, U = uint8_t, V = int8_t
  Operation: ckd_sub(0, -1)
  Vector indices: i = 0, j = 7

I'm not sure if it's even worth trying to find out why this happens, since my fork tests correctly on all platforms with all implementations. I could upstream all my changes if you agree to having the reserved identifiers changed to allowed identifiers, ckd_* being treated as a "namespace" for some additional names, east const and clang-formatted source.