[OpenBLAS_jll] Update to v0.3.27 for 1.10

[imciner2]

This updates 1.10 to use OpenBLAS 0.3.27, which should fix at least one issue reported by a user (JuliaLang/LinearAlgebra.jl#1406). 0.3.27 is the same version used in 1.11 (#54237). I did not update all the way to 0.3.29 (the most recent we have in 1.12 and master) because that drops libgfortran 3&4 for PowerPC (due to needing GCC 11 to build).

I believe I updated all the checksums/files to make this work, specifically I did

• Updated version number in stdlib/OpenBLAS_jll/Project.toml
• Updated version number and sha in deps/openblas.version
• Refresh checksums by running make -f contrib/refresh_checksums.mk -j openblas

cc @giordano, @ViralBShah

[ViralBShah]

I suspect an openblas bump on lts will not be accepted. We will need to backport a patch.

Wonder what @KristofferC has to say.

[imciner2]

Hmm, looks like someone bumped the compat bound to 1.11 for 0.3.27 when it was updated https://github.com/JuliaRegistries/General/blob/fb40294a9aa1fd3dd26ec359bc11a03a6db58987/jll/O/OpenBLAS_jll/Compat.toml#L23.

We will need to backport a patch.

The problem with this is figuring out what has changed to fix it, and if it is even in a state we could apply onto 0.3.23. Because at least a quick glance into the OpenBLAS code base doesn't show any obvious patches that might have done this.

[gbaraldi]

I don't think it's that big of deal to backport jll bumps. Specially if they aren't breaking

[KristofferC]

I personally think it is fine.

[imciner2]

I did some more digging, and there is a choice now. Either pull 0.3.27 back to 1.10 (which this PR does), or it looks like backporting just commit OpenMathLib/OpenBLAS@9019bc4 onto our 0.3.23 build will fix the issue that was reported in JuliaLang/LinearAlgebra.jl#1406.

[ViralBShah]

I feel like BLAS upgrade could have other issues with reproducibility. May be best to simply apply the patch.

[ViralBShah]

We do have a patched 0.2.23 now.

[imciner2]

Replaced with #59346 which only pulls the two needed patches.