Add support for convolutional Parseval regularizers

[Obi-Wan]

Description

In CEA-MetroCarac/auto-denoise#11 we introduced the machinery for learning convolutional Parseval regularizers from data. They are data-driven convex regularizers that can deliver higher reconstruction quality even when just using a simple l1-norm potential. Here, we introduce support for using sets of these filters as regularizers.

TODO

• Implement new feature:
  • Add convolutional operator for filterbanks
  • Add regularizer that uses the convolutional operator
  • Use the more generic convolution object for the single-filter convolution operator
• Add or update unittests.
• Add or update docstrings.
• Add or update documentation.

Notes

These convolution operators are introduced in:

• M. Unser and S. Ducotterd, “Universal Architectures for the Learning of Polyhedral Norms and Convex Regularizers,” SIAM J. Imaging Sci., vol. 19, no. 2, pp. 1272–1299, Jun. 2026, doi: 10.1137/25M176297X.