curve_fit

[ev-br]

I know you're not a fan :-).

Still, we'd need to add the new bright and shiny algorithms to what is now scipy.optimize.curve_fit. It seems to me it only depends on the least_squares interface, and the latter is not very likely to change substantially. Hence there's no harm in making a branch from the NLSQ PR, adding the curve_fit (one commit, or two), and submitting that.
For starters, I'd keep the loss functions out --- just add bounds and method kwargs.

[nmayorov]

I will do it soon then, with loss option from the start.