Cannot compute incomplete gamma function for some arguments

[araujoms]

Essentially the same issue as JuliaMath/DoubleFloats.jl#183

If I call gamma(Float128(a),x) for x >= 3 I get an error, because gamma goes to a branch where it uses the function logabsgamma, which is not defined for Float128. This comes from the deprecation of lgamma and its replacement with loggamma and logabsgamma.

In DoubleFloats I fixed this by defining the function in terms of its BigFloat version. I could do the same here or, if you prefer, define it in terms of the Float128 version of lgamma with something like

function logabsgamma(x::Float128)
    sign = x >= 0 ? 1 : 2*mod(ceil(Int64,x),2)-1
    return lgamma(x), sign
end

It would also be nice to define a loggamma function with the same behaviour of the SpecialFunctions one for consistency.

[RalphAS]

The version using only Float128 is preferable, and I agree that we should make the related API compatible with SpecialFunctions.