Refactor algorithms module

[Obi-Wan]

Description

The algorithms module currently suffers from a list of different problems:

1. It is quite large, making browsing/changes a bit too tedious and error prone
2. The N2N and DIP algorithms have lower reproducibility, because the training/cross-validation pixel mask is randomly re-initialized at each call of the training function
3. Also for N2N and DIP, the current infer method expects input data to be already structured in a way that can be almost directly fed into the model. However, there is not function to prepare the data for more complex data structures/organizations
4. The API is a bit all over the place, differing widely from method to method

This PR proposes a set of changes that address those points by:

• Splitting the algorithms module into a sub-package, addressing point 1
• Adding a data preparation function for the N2N and DIP algorithms, which addresses points 2 and 3
• Making the training API more homogeneous, addressing point 4

TODO

• Implement changes:
  • Split the algorithms in different files
  • Homogenize training API
  • Adding data preparation function for N2N and DIP
• Add/update unit tests
• Add/update documentation:
  • Examples
  • Docstrings
  • Tutorials

Notes

This is a prerequisite for future enhancements like:

• 2.5D data treatment for N2N
• 1D and 3D data treatment for all algorithms
• Support of data transforms to DIP, which would enable tackling problems like deconvolution, super-resolution, and other inverse problems like tomography, phase retrieval, etc.