Replace msf_gif submodule with single header (–711 MiB) + macOS build fix

[imsys]

Saves 711 MiB of disk space by adding the msf_gif header file instead of the whole repo that contains mostly test cases.
It saves 354 MiB of bandwidth, but it is stored twice, in .git/modules/vendor/msf_gif and vendor/msf_gif, so that's why it takes double the size.

The full TIC80 repo after getting all the submodules, before making any build takes 1203.7 MiB, so this lib alone was taking 59% of disk space. I verified and it was due to many test cases that are unnecessary to the TIC80 project.

[imsys]

Discussed in the previous pull request #2830

[imsys]

During the first build, the HTML build job failed with the error:
OSError: [Errno 39] Directory not empty: 'joystick'

Github copilot says it was a bad luck during the cmake --parallel and suggested to remove the parallel.
https://github.com/copilot/share/8215410e-4be0-80c2-9812-94098046484e

Maybe by removing the msf_gif repo makes a faster build, that increases the chances of parallel build errors? idk.
Maybe it was just bad luck and it's still work keeping the parallel.

But we keep this in mind in case we stumble upon the same problem.

[nesbox]

Thank you, I think you need to add yourself to the credits in readme.md in one of the future pull requests.

[imsys]

Sure, but I’ll send a few more contributions before adding to the credits. :)
I’m kind of on vacation, and I plan to dedicate about two weeks to the TIC-80 project.
My goal is to fix most of the games that were broken by regressions. I think I’ve identified about three or four different cases (or causes).

After that, there’s a very difficult task that I’m not completely sure I can solve. Some games run too slowly on cheap retro handhelds, like the Emuurom and others. I’m not sure if the issue comes from the libretro core specifically, or if the standalone version on those devices has the same slowdown. Either way, I plan to investigate and try to improve performance.