make functions usable with ForwardDiff.jl

[tpapp]

While ForwardDiff.jl recommends that functions accept generic Vector, StatsBase.jl specializes functions to RealArray, making them unusable with ForwardDiff.jl (eg mean_and_var, etc).

I would be happy to contribute a PR, but I am wondering what the preferred solution would be, eg

1. extend RealArray to a union with types that ForwardDiff.jl would use,
2. replace RealArray with Vector

[andreasnoack]

I don't that we need to be overly strict about the typing of these functions. We should consider using AbstractVector and a PR would be great.

[tpapp]

Apparently this is fixed by JuliaDiff/ForwardDiff.jl@934b3b2 which makes Dual{N,T<:Real} <: Real.

If you think it is necessary, I can still submit some tests to make sure that this keeps working.

[nalimilan]

Tests are certainly a good idea if you want some functionality to keep working in the future.