406 hades weekly fail - hotfix

[egillax]

Fixes #406

The issue was that in scikit-learn 1.3 they added a new feature to decision trees to deal with missing data. This broke our serialization of scikit-learn models to json. The reason this was not detected in our regular tests but in the weekly Hades tests is that there was inconsistency in how python environments were set up for testing. It was hardcoded in our regular test github actions to use python 3.7, which has gone out of support. This resulted in our tests using an older version of scikit-learn which didn't have this issue.

In this PR I've

• Fixed the serialization of trees so they work both with scikit-learn 1.3 and older versions.
• Simplified the actions so the setup of python env uses configurePython function for consistency
• Update ubuntu runner to use 22.04 which is the latest long-term supported release. This was necessary since otherwise a system library needed to be updated for scipy to work correctly in python 3.11
• Update README to mention currently supported python versions 3.8 and newer are supported

Now the tests test against the default python version in conda, which at this moment is python 3.11. Other option would be to hardcode somehere a different python version to test against. But I'd expect versions 3.8-3.11 to work currently.

@jreps could you review/approve this?