Fix pytorch cpu jupyter server rock

[misohu]

This PR fixes (refer to the issue for the whole discussion):
canonical/data-science-stack#47

This PR also fixes some problems from here #78. Namely it updates python, miniforge and it uses mamba. We still need to figure out the s6 and builds for different architectures.

I updated the rock based of upstream changes in branch v1.8-branch here

Key changes:

• Python 3.11
• Miniforge 23.3.1-1
• Torch 2.1
• Using Mamba for installing some ML tools
• adding mlflow 2.1. lib

Changes in tests:

• now we test for mlflow but also for scikit learn and scikit learn image
• fastai is no longer part of torch its a separate instalaltion which no longer part of upstream Dockerfile
• kfserving lib is also no longer part of upstream image

CI is failing because trivy scan times out as the rock is too big. I fixed this in this PR which should be merged first and then we can rerun ci (I did that in previous commit with pinned branch in CI).

[orfeas-k]

I 'm adding some comments here since I could not comment on unchanged code:

• Is there a chance we need to update the following line? Although I see they still set the same NB_USER .

chown -R 584792:users home/jovyan

• I also noticed that upstream has removed that line and modified the one below as seen here.

[orfeas-k]

Approved! Good job @misohu. This was a convoluted ROCK actually with a lot of digging needed but you pulled through as always 👏