Improve parameter tuning

[Obi-Wan]

Description

Hyperparameter tuning is fundamental for obtaining good reconstructions when using regularizations. corrct currently offers support for single hyperparameter tuning, based on the L-curve and cross-validation methods. However, with growing needs for other work distribution tools than the basic ThreadPoolExecutor, the current API and infrastructure are inadequate to support further developments. This PR cleans the API, while making it more flexible and transparent. It also refactors internal code to use a more structured and scalable approach (using simpler functions and fewer nested class methods). Finally, this PR also introduces simple support for K-fold cross-validation and tracking performance metrics of the reconstructions.

TODO

• Implement new feature:
  • Refactor naming: spawn -> init, and call -> exec (while keeping the old deprecated interface)
  • Generalize parallel and serial processing code
  • Add support for passing keyword arguments to both init and exec functions
  • Add support for specifying the mask argument name in the cross-validation processing
  • Add support for tracking performance metrics of the individual task and the batches
  • Add support for k-fold cross-validation
  • Optional: add support for no init function
  • Optional: add support for variadic return arguments for the init function
• Add or update unittests.
• Add or update docstrings.
• Add or update documentation.

Notes

Performance metrics and all reconstructions are currently only returned when requested with the return_all argument set to True. This should reduce memory usage when those features are not necessary.

[Obi-Wan]

We should probably use this PR to also add decent documentation for this feature.