TRF vs MINPACK performance on unconstrained problems

[nmayorov]

Despite being methodologically the same, TRF usually comes short of MINPACK by a few number of function evaluations required in unconstrained problems.

I don't think it's necessary to adopt all logic from MINPACK, but it is worth trying to find a crucial part which brings MINPACK ahead. The number one suspect is trust radius shrinkage strategy.

But, nota bene, that the main application of TRF is constrained problems and adopting more logic from MINPACK will not necessary improve this part, and may even degrade it. Need to be careful in this regard.