Add option for spectral gradient

[f-brinkmann]

When going though the paper I wondered if we should add an option to use the spectral gradient for backwards compatability with the AMT version.

[robaru]

Good point...

The spectral gradient was something that I did not focus on since we never saw much of a difference when using broadband sounds. The gradient happens to work better than amplitudes only for rippled spectra sources.

Let's include it later for back compatibility.

[robaru]

added in fitting/spectral_gradient branch. IMPORTANT: this branch is based on the branch fitting which will need major rework after integration of pf.Coords into develop

[robaru]

The stub implementation is already in place via the Barumerli2023pge class in auditory_representation.py. This class is registered and ready for implementation—both __post_init__ and sigma_matrix are prepared to receive the spectral gradient computation once it's developed.