Scout: better loss function

[jsappl]

Relax to squared Frobenius norm like https://arxiv.org/pdf/2305.16432?

[jsappl]

I'm quite happy now and getting some good results with approximating the matrix inverse:

$$\lVert{MA-I\rVert} \to \text{min}$$

instead of computing the expensive SVD for the condition number $\kappa(MA)$ as a loss function