include energy kernel for MMD loss

[florence-bockting]

Situation/Background:
For the MMD loss only the Gaussian kernel and the inverse multiquadratic kernel are implemented
Changes made:
I implemented additionally the energy kernel as suggested in [1,2]
Still ToDo:
Sofar the kernel for the MMD loss can't be selected by the user but is by default the "gaussian" kernel.

[1] Feydy, J., Séjourné, T., Vialard, F. X., Amari, S. I., Trouvé, A., & Peyré, G. (2019, April). Interpolating between optimal transport and mmd using sinkhorn divergences. In The 22nd International Conference on Artificial Intelligence and Statistics (pp. 2681-2690). PMLR.
[2] Feydy, J. Geometric Data Analysis, beyond Convolutions. Ph.D. Thesis, Université Paris-Saclay, Gif-sur-Yvette, France, 2020.

[paul-buerkner]

Thank you for this PR! As discussed offline, I will close the PR as it was made against master rather than dev.