Make training step re-runnable

[arthurBarthe]

This corrects for minor issues that were occurring when running the training step. It closes #33. Note that the bug in #42 does not seem to occur as long as the processed data is large enough.

[raehik]

Rebased from main: CI changes from #56 , some project meta changes from 630efad

[mondus]

We want to get input from @CemGultekin1 on if our changes to mlflow commands are reasonable. We would like to ensure that this PR incorporates the changes that were made in the informal fork. The question is "Have we incorporated the same idea?", "Can our mlflow commands be improved". We certainly need to document better when the arguments are.

[dorchard]

Another query: the example command line includes -P exp_id=<data_experiment_id> -P run_id=<data_run_id> How important is this and what guidance should the user be given about how to chose these? I found that they had to be filled in, but it looks like there is "free choice"?

[raehik]

#73 covers these changes -- closing.