Fractional polynomial model

[choisy]

Here I would make 2 major changes:

• combine find_best_fp_powers() and fp_model() into one single function fp_model() the behavior of which depends on the arguments values that are given (either specific values for estimating a specific model or a range of values to look for (and estimate) the best model over that range). The rational is that we want the user experience as seamless as possible.
• I don't think we want to feed the function with a sequence p of powers to brute-force try in turn. Instead, we want to find the best power value in a range using an optimization algorithm as implemented in the optim() function (unless you manage to convince me of a particular application case where exhaustively looking at predefined values of power could be useful).

[choisy]

I would make the exact same comment for the functions plot_gcv() and lp_model().