Fix bug with typed_fad_gradient!()

[utkinis]

Current implementation of typed_fad_gradient!() contains a bug related to access to non-existent field of a structure and is inconsistent with typed_fad_gradient(). When trying to call forwarddiff_jacobian!() of a function that returns Vector, an error occurs. In this example:

using ForwardDiff

F(x) = [x[1]^2, x[1]*x[2], x[2]^2]

J! = forwarddiff_jacobian!(F, Float64, fadtype=:typed)

A = zeros(3, 2)
J!([1., 1.], A)

after execution got

`g!` has no method matching g!(::Array{Float64,1}, ::Array{Float64,2})

If try to make A=zeros(6) got

type Array has no field g

After that fix, one can use Matrix of the same shape as in the output of forwarddiff_jacobian().

[papamarkou]

Thanks for the fix @krjackvinator, I will look at this in a few days as I am currently travelling.

[papamarkou]

@krjackvinator apologies for taking me a while to look at this, do you want to fix your PR so that it doesn't fail, and then I can check it and merge it?

[utkinis]

@Scidom i apologize for not replying for so long time(i had been working on my graduate thesis). I had to change tests a bit for forwarddiff_jacobian!() because this function should take type Array{Float64, 2} instead of Array{Float64, 1}(it doesn't know if argument function returns one value or array of Float64). Unfortunately, now this PR contains merge conflicts. I could merge master with my branch, but then commits from master will get into PR, so i'd better wait for your decision.

[jrevels]

This now fixed on master as of the merging of #27.