add manylinux2014 build targets for linux

[mattip]

we have manylinux1, manylinux2010, but not manylinux2014 x86 tarballs.

xref matthew-brett/multibuild#355

At some point we could rename these to not use manylinux tags to save the triple builds.

[mattip]

Seems appveyor is failing due to this issue msys2/MSYS2-packages#2058. Tangled web indeed. Luckily , appveyor is not needed for this PR which only involves linux

[tylerjereddy]

Any reason to update the OpenBLAS submodule? I thought we were pulling in the version/hash based on BUILD_COMMIT?

[mattip]

Any reason to update the OpenBLAS submodule

No, that was a mistake, thanks.

[tylerjereddy]

Alright, I'll put this in.

I believe NumPy is using an even newer version of OpenBLAS in openblas_support.py (the longer version suggests you needed a patch after the stable release) so the hash regenerations for v0.3.10 may not be too disruptive (although if you do a point release of NumPy with an older version of openblas_support.py it is possible that NumPy wheel builds might need a backport update to the hashes).

SciPy is still on OpenBLAS v0.3.9 so should not be disruptive there.

[tylerjereddy]

Thanks @mattip

[radarhere]

So I presume this won't actually appear at https://anaconda.org/multibuild-wheels-staging/openblas-libs/files (without a hash) until the next version of OpenBLAS is released. Just asking, is there an estimate for when that would be?

[mattip]

The newer one already appears https://anaconda.org/multibuild-wheels-staging/openblas-libs/v0.3.9-452-g349b722d/download/openblas-v0.3.9-452-g349b722d-manylinux2014_x86_64.tar.gz. For matthew-brett/multibuild#355 we should also rebuild v0.3.10, which I thought this PR would do.

[mattip]

@radarhere I created the v0.3.10 manylinux2014 tarballs as well. Thanks for pointing it out.