Fix for dual numbers

[Expander]

See issue #40

[NAThompson]

@Expander : This looks great thanks!

One more thing just to perhaps head off future problems: I found this because I was trying to autodiff some code, so eventually I want to get derivatives.

I wonder if we could add a unit test for the derivatives computed via ForwardDiff? To wit, maybe use $z \frac{\partial Li_s}{\partial_z} = Li_{s-1}(z)$?

[Expander]

Good point, but there are such unit test already. :)

To be precise, there are tests for the real functions li0, reli1, reli2, reli3, reli4 and the general reli. To find the tests, just grep ForwardDiff test/*.jl. They have been implemented by @longemen3000 when he added the support for ForwardDiff and ChainRules.

Back then, @longemen3000 said that ForwardDiff does support derivatives of (holomorphic) complex functions. When this is added, the complex polylogarithms will get derivative support as well (including tests).