Remove MLFlow framework

[jatkinson1000]

We probably don't want MLFlow for distribution?

Easy to remove the entry point.
Harder to deal with the way datasets are currently saved and then re-opened.

[jatkinson1000]

Perhaps provide documentation for how a user could build an MLFlow framework around the final package.

Discussion with @arthurBarthe Will need to change where processed datasets are saved in the code.

[arthurBarthe]

I agree that the MLflow component should not be compulsory. We might want to make it easy to interact with it, possibly through some additional API, but we can leave that for later.

[jatkinson1000]

Files that deal with MLFlow:

• cmip26.py - saves dataset in mlflow framework - need to save elsewhere.
• trainscript.py - uses MLFlow to log run parameters. Can probably be removed easily, but consider logging elsewhere?
• data/ - utils.py and readData.py both use mlflow to load data and log?
• listdata.py - used to display MLFlow runs. Can probably be removed.
• MLproject - defines entry points to scripts. Rewrite to json inputs and reduce argparsing?
• analysis/ - contains some files that load mlflow data so would need refactoring.
• mlruns - stores runs and associated data. Not currently part of repo as in .gitignore (also means various jupyter notebooks are unusable as data not there).
• Various jupyter notebooks use mlflow to load data.

[mondus]

For a first release it is suggested that we leave MLFlow within the code and this can be refactored later if required.

[MarionBWeinzierl]

#85 and #95 are introducing a CLI for the data and training step, respectively. The inference step is still to be done, and the Jupyter notebooks have to also use those changes and don't yet (see #82).

The refactoring work is done/merged on the dev branch.

[raehik]

#97 removes MLflow integration.