fdfitter: Improve inversion performance (Friday afternoon)

[JoepVanlier]

Model inversion can be pretty slow and generally, it is cumbersome to have to specify the inversion bounds.

This draft fixes both these issues by introducing a new inversion mode that does simple rootfinding to find the outer extents of the to-be-inverted range. As a fallback method, we fall back to the old method of using least_squares to invert the function.

Differences with the old method:

• Default bounds make sense for F, d models. While arguably, this makes the Fitter less general, it these bounds do hold for any sort of F, d model you'd throw at it (positive only).
• Twistable wlc makes use of the rootfinding method (I'd say this is optional, we might hold off on this until 2.0, since technically, it is breaking, but only by a very small amount).
• Small bugfix in the bounds of the interpolation method.
• Pass a method flag instead of interpolate=True.