Wavelet transforms support

[Obi-Wan]

Description / Context

Wavelet transforms enable representing signals in spaces where they are often sparser. This can be leveraged for imposing prior knowledge on the reconstructions, or as up-scaling / down-scaling functions in multi-scale analysis.
This PR implements a wrapper for ptwt (PyTorch-enabled version of pywt) for the discrete wavelet transform and its inverse (i.e., wavedecn and waverecn), and for the stationary wavelet transform (i.e., swtn). The inverse stationary wavelet transform might be added later.

TODO

• Implement feature:
  • Implement wavedecn and waverecn wrappers for existing 1D, 2D, and 3D functions.
  • Implement swtn
  • Optional: implement iswtn
  • Convert the SWTN loss to use the new swtn function
  • Implement a DWTN loss to use the new dwtn function
• Add/update documentation:
  • Docstrings
  • An example of different regularizations
• Add/update examples (of different regularizations)
• Add/update tests

Notes

This code is heavily inspired by ptwt: It borrows and generalizes various of its internal functions.

[Obi-Wan]

This branch is now complete. The iswtn function can wait, as it is not going to be used in the near future.