Tracking issue for CB3 migration

[jaimergp]

I am taking over the excellent work @Lnaden started in this other PR #177

Pending issues

• OpenMM does not build - fixed in Upgrade to CB3 & Azure - part two #180
• Add the BINSTAR Token to the Azure org to handle uploads - fixed in Upgrade to CB3 & Azure - part two #180
• Make sure we are uploading to the correct channel/label
• Linux: mount repo to $HOME instead of /io so we get the conda_build_config.yaml pinnings.
• Update README with new changes
• Build OpenMM docs. Shall we create a separate package openmm-docs?

Further work and ideas

• Figure out how to unify recipes for main / beta / nightly builds
• Idea: Investigate improved CF compatibility through conda-forge-pinning compatibility
• Idea: openmm-docs can be created as a separate package, simplifying the Docker images and reducing build time.
• PowerPC builds should be possible too once the CF PPC Docker images are available (WIP).

Changelog

• 2020.01.17. Nightly uploads should work now after Fix uploads #182
• 2020.01.14. Docker images are now living in omnia-md/cf-docker images and being uploaded to omniamd/linux-anvil-cuda in Docker Hub. In principle, everything is setup for tonight's nightly build, which should result in uploads to omnia-dev.
• 2020.01.13. Docker images with TeXLive had an issue with glibc that resulted in core dumps at install time for OpenMM. They have been replaced with the Conda Forge CUDA images, augmented with more CUDA versions (8.0, 9.0, 9.1). The fork resides in https://github.com/jaimergp/cf-docker-images. They have recently added ppc64le support for their CUDA images so we could build those too.