Move CI to GitHub Actions

[Morwenn]

I planned to move the continuous integration to GitHub Actions at some point, but the recent changes to Travis CI mean that we will likely have to move sooner rather than later.

[Morwenn]

Forgot to mention it, but commit 1de7526fc01d0f173d18d0ea93f55af57b850d38 add a workflow for Ubuntu.

[Morwenn]

I got the following kinds of builds working:

• Debug, Release
• Ubuntu, MacOS, Windows
• g++ (including MinGW-w64), clang++ (including AppleClang)
• Valgrind, asan, ubsan
• libstdc++ and libc++ (I think)

While a bit different from the old Travis CI builds, I believe that I cover an equivalent amount of possibilities. There main differences are as follows:

• No Clang build older than Clang 6
• GCC builds for MacOS
• Asan and ubsan builds for MacOS
• No Valgrind builds for MacOS

This is pretty satisfying as everything runs smoothly. There are a few additional steps before closing the issue:

• [x] Deactivate the Travis builds for cpp-sort
• [x] Remove .travis.yml
• [x] Remove the OSX suppression file since I don't have Valgrind build on MacOS anymore
• [x] Remove the mtime_cache_globs.txt file, not used anymore
• [x] Change the README to mention the new supported (as in tested) compilers
• [x] New README badges for the builds, move them to "Compiler support & tooling"
• [x] Use strategy.matrix.config to simplify build YAMLs
• [ ] Re-enable ccache through actions/cache, I don't think that mtime_cache is needed anymore

[Morwenn]

Commits 2cc9016a0830f70906b98ee194514ce75176dc89 and 13157e2aa1fccc43d5eb0e7dffd40f431de64b71 simplify the MacOS and Ubuntu build matrices respectively, making them more understandable, less redundant, and easier to maintain.

[Morwenn]

actions/cache@v2 has some significant drawbacks so far:

• No way to clean the cache through the UI
• Rather verbose
• Doesn't play well with ccache so far, it still lacks critical features for that

Considering that the build times are currently decent even with a full rebuild, I consider that making the workflows more complicated is not worth my time. I might give it another try in the future once features and usages have evolved, but currently it's got more cons than pros.

As such, I consider that this issue is in a good enough state for now, and ready to ship.