Apple Silicon support

[lutzroeder]

Building arm64 binaries:

• Apple Silicon Universal Support electron-userland/electron-builder#5475
• macOS default architecture electron-userland/electron-builder#5495

Enabling build tools and scripts:

• Please provide universal2 wheels on macOS numpy/numpy#18143
• M1 release and nightly binaries pytorch/pytorch#51854
• trouble installing on apple silicon M1 scipy/scipy#13409
• Native wheels for the macos/arm64 platform (Apple Silicon M1 hardware) scikit-learn/scikit-learn#19137
• M1 arm64 release binaries tensorflow/tensorflow#47782

[ogrisel]

Note: scikit-learn/scikit-learn#18861 is about pip refusing to install the macosx_10_9 on Intel Big Sur. It is unrelated to Apple Silicon support (different hardware) unless you plan to rely on Rosetta 2 (the Intel / ARM translation layer of macOS).

We do not yet publish Apple Silicon compatible wheels on pypi.org. However, conda-forge has already native binary packages for Apple Silicon (known as the osx-arm64 platform for conda-forge): https://github.com/conda-forge/miniforge#download

This includes numpy, scipy and scikit-learn among other things.

[lutzroeder]

@ogrisel yes, this is needed at build-time only and running through Rosetta is fine for now.