Pin python version in a strict way

[mathbunnyru]

Fix: #1352

This is another approach to fix the python version (mamba doesn't change the python version in #1351).

For now, I'm not 100% sure about this PR, because @parente wanted the python patch version to be floating (#778 (comment)).

I want to see how it's gonna work for pyspark image, which was updating python version, even it might not be needed.
Also, it's nice to see the same tags in all the images and to control python version via a variable (it was unexpected for me that two images had different python versions at all).

[mathbunnyru]

Seems to work fine.
This PR:
https://github.com/jupyter/docker-stacks/pull/1354/checks?check_run_id=2722386028#step:5:6153
Built image size: 3.32GB
jupyterlab_server and python versions stay the same

Master:
https://github.com/jupyter/docker-stacks/runs/2719734258#step:5:6155
Built image size: 3.41GB

Also saves 90 MB of information.

[consideRatio]

This is too technical for me to parse, can you describe what is changed in what was previously written to the pinned file vs what is now written to the pinned file?

[mathbunnyru]

For example, the current version of python is 3.9.2.
Currently, in the master branch we will pin the version like this python 3.9.*.
So, during conda update (and even install), we might update (and even downgrade) python version to 3.9.x.

The change: write exactly python 3.9.2. This way we pin python versions in a strict way and it won't change during conda update / install.

[consideRatio]

Excellent thank you!

[consideRatio]

Questions in my mind

• Are we in agreement to want to pin our Python version while installing the docker image?
  • I think we should enforce a single python version to be used across all our build steps, and if we need to rely on adding a line to the pinned file when using mamba, then I support that at least during the build.
• Should we have an entry in the $CONDA_DIR/conda-meta/pinned file after our build's various updates to conda packages are complete?
  • If so, how tight should that version pin be?

[mathbunnyru]

Questions in my mind

• Are we in agreement to want to pin our Python version while installing the docker image?

  • I think we should enforce a single python version to be used across all our build steps, and if we need to rely on adding a line to the pinned file when using mamba, then I support that at least during the build.

As this review shows, we can rely on adding a line to the pinned file (python version is not changing).

• Should we have an entry in the $CONDA_DIR/conda-meta/pinned file after our build's various updates to conda packages are complete?

  • If so, how tight should that version pin be?

I think, it's ok to pin the version in a strict way - we provide our docker images with complete python tag like jupyter/all-spark-notebook:python-3.9.4, so it won't be obvious for users that they might change the python version after some conda installs / updates.

[mathbunnyru]

And, I think we're updating quite a lot our mambaforge installation (at least we try), so we will be always with quite recent version of python).

[mathbunnyru]

Let's try this one out.
Now, after switching to mamba, the python version is not changing.
But let's try to merge this to see how it goes for the users. It's easy to revert in case something goes wrong.