Do we really need these platforms? There was 1 real user request for these platforms: #788

I am asking, because this is not free for us:

• We have to maintain all the build infra
• We have to run CI for this regularly

I think that msgspec can be compiled locally quite easily and quite efficently. So, proving wheels is not 100% required for it to work.

Moreover, there's not much platform specific code in msgspec for us to test it heavily.

Yes, this will increase the CI time, but for people who use these targets, installing the wheel will be much faster than compiling it on the spot

I guess we can compile manylinux and musslinux on different GitHub Runners

Agree with @sobolevn.

I'd also be cautious to add more platforms on a hunch. If there's a real demand for it, by which I mean at least a ticket asking for a specific platform with a good reason, I'm all for it. But otherwise, maybe let's hold off with it a bit.

In theory, without cibuildwheel, we could skip using QEMU, use zig cc instead, and compile the wheels on an x86 platform, but we would still need to test them, which takes about ~100+ seconds per wheel.

If we remove the musllinux wheels for s390x, ppc64le, and armv7l and keep what's in this PR, the build time will remain roughly the same - less than ~10 minutes per GitHub Runner.