jacobian erroneously calls finite_difference_derivative

[danielwe]

This call should go to FiniteDiff.finite_difference_jacobian:

https://github.com/JuliaComputing/NonlinearSolve.jl/blob/cd00b8e4e3302be2cc724de126f5ca184d0e252e/src/jacobian.jl#L35

The reason the tests still pass is that the solution to the test problem is proportional to the initial guess, and the test function works for scalar u. If you change the initial guess so that it's not proportional to [1.0, 1.0] the algorithm will fail to converge, and if you use a function that fails for scalar u (e.g., by indexing into it) an exception will be thrown when trying to compute the jacobian.