Paths and file management

[alex404]

So this does quite a fair bit of refactoring.

• I setup the analyses in statistics.py to return dataclass instances to make things more manageable and readable.
• I also made a push to a more pathlib style of directory management, in particular in the runner submodules.
• Generally going for a style where the at the top of the framework command (e.g. intialize, analyze or train) all necessary variables are extracted out of the DictConfig.
• Saving analysis results to disk was the original push for this, and that's done too.

I definitely wasn't perfectly complete or consistent with how I did this, but I think things are significantly better organized than they were before. Ideally this provides a good reference for how to factor/refactor such code in other parts of the library. Definitely open to feedback on review.x

Resolves #43

[alex404]

This probably seems more complicated than it is. The bulk of the changes are moving blocks of arguments around into dataclasses.

[alex404]

Actually hold on just discovered a bug

[alex404]

Okay seems to be fine now

[fabioseel]

Seems mostly fine, just want to hear your comment thoughts on my comment in plot (moving the inner function out to module scope)
and in statistics (using the defined dataclasses already within the functions)

[fabioseel]

just a minor change to avoid the duplicate code

[fabioseel]

Looks good to me now :)