Switch from Miniconda to Miniforge

[romainx]

Hello,

I'm trying to perform a structuring change by switching from Miniconda installer to Miniforge.

This repository holds a minimal installer for conda specific to conda-forge.

Here are the main reasons

• Switch to open source since Anaconda has recently updated its TOS to restrict commercial usage of its repositories (including the repo hosting Miniconda installers) -> This could impact users of Jupyter Docker images
• It's built to use conda-forge channel by default
  • it is the channel we are using that is not restricted
  • there is no need to prepend it anymore
  • other default channels that are concerned by the new TOS are not configured
• Support various CPU architecture (installer can be adapted through ARGS), could be a good asset to build multi-arch images.
• Support for pypy by using alternative installers Miniforge-pypy3 (thanks @mathbunnyru)

Some minor pros

• There is almost no change in the installation process
• Base image size seems to identical or slightly reduced
• Speed up the package dependencies resolution (only one channel)

Implementation

• Installation is pretty much the same 😊
• Split the installer definition in several ARGs to ease overriding and for better clarity (I hope)
• Removed the environment variable MINICONDA_VERSION because I'm not sure it makes sense to replace it by the Miniforge equivalent
• Removed the strict channel priority of the conda installer because
  • it is no more necessary with only one channel
  • it makes installation of packages from other channels (set by users) harder
• Decision to keep the installer checksum as an ARG instead of downloading it for security reasons (@mathbunnyru suggestion)
• Documented in "Common Features" the way to install packages from defaults channels

Notes on multi-arch images

Here is a table displaying the various CPU architecture available

Ubuntu	Miniforge	Miniconda
arm64	linux-aarch64	-
amd64	linux-x86_64	Linux-x86_64.sh
ppc64le	linux-ppc64le	Linux-ppc64le.sh

List of ubuntu:focal architectures

curl -s https://hub.docker.com/v2/repositories/library/ubuntu/tags | \
    jq -r '.results[] | select(.name=="focal") | .images[].architecture'

# arm64
# s390x
# amd64
# ppc64le
# arm

[mathbunnyru]

Another great advantage is an ability to use pypy - maybe someone wants it, because it might give good speed improvement.

[mathbunnyru]

LGTM

[parente]

This also looks very good to me, @romainx. Thanks for staying on top of conda developments. Merging now. We can deal with issues as they arise.