Need to bound nonlinear solver parameter ```alpha```

[mptino]

alpha is currently calculated as

alpha = 0.5 * max(1., x_prev.Norm()) / f.Norm()

but this can lead to over-estimations in the magnitude of alpha.

For example, one specific problem initialized alpha as alpha = 4.8, but the iterations experience severe error oscillation (low error $\longleftrightarrow$ high error) and fails to converge within reasonable number of iterations.

However, when choosing alpha ~ 1., we obtain rapid convergence (less than 20 iterations).

Need to investigate acceptable bounds for alpha although currently none exist in their implementations in scipy/optimize/_nonlin.py.

[nasserma]

Addressed with commit 1913c43.