Fix osx-arm64 builds failing on missing openmp pin

[h-vetinari]

This happens since the openmp 12.0.1 builds, with

   File "/Users/runner/miniforge3/conda-bld/scipy_1628206424706/_build_env/venv/lib/python3.8/site-packages/omp/__init__.py", line 107, in init_not_msvc
    self.libomp = ctypes.CDLL(libgomp_path)
  File "/Users/runner/miniforge3/conda-bld/scipy_1628206424706/_build_env/lib/python3.8/ctypes/__init__.py", line 373, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: dlopen(/Users/runner/miniforge3/conda-bld/[...]/bin/../lib/libgomp.dylib, 6): no suitable image found.  Did find:
    /Users/runner/miniforge3/conda-bld/[...]/bin/../lib/libgomp.dylib: mach-o, but wrong architecture
    /Users/runner/miniforge3/conda-bld/[...]/lib/libomp.dylib: mach-o, but wrong architecture
Preparing wheel metadata: finished with status 'error'

Additionally, the pin from conda-forge/conda-forge-pinning-feedstock#1727 isn't taking here, because scipy does not depend on llvm-openmp, so I guess it's coming in through a dependency that hasn't been rebuilt with the pin yet.

Should we add an llvm-openmp constraint for osx-arm here, or how best to tackle this?

[isuruf]

Should we add an llvm-openmp constraint for osx-arm here, or how best to tackle this?

It doesn't matter as scipy is not linking to openmp

[isuruf]

See serge-sans-paille/pythran#1868

[h-vetinari]

Thanks!

[h-vetinari]

Do you already want to rebuild the pythran feedstock with that patch or wait for serge-sans-paille/pythran#1868 to be merged?

[h-vetinari]

This has been fixed by #179