Vector-Valued Nonlinear Functions

[pulsipher]

In connection with #105, as we better decouple the transformation/solution schemes from JuMP we should enable InfiniteModels to handle vector-valued nonlinear functions. This should equate to an extension of our existing nonlinear interface and should be possible to implement in a nonbreaking way.

[pulsipher]

Now that InfiniteOpt relies on JuMP's nonlinear interface, I am closing this. However, adding vector-valued nonlinear functions is on JuMP's development roadmap now.