Refactoring of the MLIP modules and other non-urgent things

[QuantumChemist]

I want to do a code refactoring and restructure some Makers and the MLIP modules, also check VASP Maker and custodian settings, plus improve the unit test setup (like using automate2 mock_vasp direct import, match unit tests to src structure).

This PR might take a while and will not change anything in the current code functionalities or features.

I am just collecting all issues that don't alter the functionality or features of the code under the umbrella of this PR. Tasks might have to be split up into several smaller PRs over time

current ToDos:

• restructure MLIP input
• refactor some Makers' make() parameters
• update documentation with input changes
• import mock_vasp via atomate2 monkeypatch
• match unit tests to src structure
• remove from __future__ import annotations from all files
• rename rand_stuc labels and suffixes to rattled to distinguish it from the RSS part
• better storing of the different xyz files
• adjust the fit_kwargs handling in the docs
• add to doc how to change MLIPPhononMaker settings (e.g. benchmark_kwargs={"relax_maker_kwargs": {"relax_cell": False, "relax_kwargs": "other_kwargs"}}

final ToDo:

• check if documentation is up-to-date with all changes

[QuantumChemist]

now I need to adjust the fit_kwargs handling in the docs

[QuantumChemist]

The PR is ready to be reviewed, just need to make the last adjustments to the docs

[QuantumChemist]

I would say that this PR is ready to be merged, if there is no other suggestion for improvement, @JaGeo 😃

[JaGeo]

What preprocessing happens and would the whole fit work without it?

[QuantumChemist]

Data preprocessing sorts out data points with large forces. It's not implemented for anything else than GAP, but it will also divide the dataset into the different data types (phonon, rattled). Which means that most of the time, nothing happens during this step.

[JaGeo]

But does the workflow run without it?

[QuantumChemist]

The fit won't automatically work without it because in this step the data is also divided up into train and test set, so one would need to provide these files.