openblas: improvements to 0.3.28 patch files

[rgommers]

These changes are what I needed to use an openblas subproject with SciPy. A few bits of minor maintenance (de-duplicating configure checks, silencing build warnings, fixing some typos), and a few bugs that were caught by the SciPy test suite, which is quite a bit more extensive than OpenBLAS's own tests.

Follows up on gh-1734.

Cc @HaoZeke for visibility.

[rgommers]

I was confused for a moment about why Alpine is failing here while it was passing on gh-1734 - in addition to the Windows builds, which were failing there already. The reason is that the build configs have changed: the Alpine job in gh-1734 was using x86-64, while now there are four Alpine configs for x64, riscv64, ppc64le, armv7. The SIMD kernels for those platforms are not wired up at all, this wrap only supports x86-64 and arm64 right now. I'll update to emit a more informative error at the start of the build.

Other than that, can I just ignore the Alpine failure (for the reason as given in #1734 (comment)) @eli-schwartz?

[eli-schwartz]

Yes -- please do ignore and defer those new errors to be addressed at some future date.

[rgommers]

Thanks. This should be good to go then. It passes the relevant CI jobs here, as well as the ones on @HaoZeke's repo (HaoZeke/OpenBLAS#30), and the SciPy test suite in my local testing.

[HaoZeke]

Thanks @rgommers