add krylov_s{0,1,2,3} which use scipy optimization.root

[bremond]

Just an example of a simple quasi-Newton solver outside of numerics. It uses scipy.optimization.root with Krylov method.
iparam[9] allows for the choice of a pre-conditioner:

• 0 no pre-conditioner
• 1 BroydenFirst
• 2 DiagBroyden
• 3 Anderson

On a few tests with Chute_4000, it seems that 0 is the best choice at this level and it performs quite faster than NSGS-AC for a precision of 1e-4 (but it is not performant for better precisions).