Add derivative of logabsgamma

[devmotion]

This PR adds a derivative for logabsgamma (basically a copy of JuliaDiff/ForwardDiff.jl#585 for Tracker). The derivative cannot be defined in DiffRules since the function returns a tuple.

Currently:

julia> Tracker.gradient(x -> logabsgamma(x)[1], rand())
ERROR: MethodError: no method matching _logabsgamma(::Tracker.TrackedReal{Float64})
...

julia> Tracker.gradient(x -> logabsgamma(x)[2], rand())
ERROR: MethodError: no method matching _logabsgamma(::Tracker.TrackedReal{Float64})
...

With this PR:

julia> Tracker.gradient(x -> logabsgamma(x)[1], rand())
(-1.808248603306234 (tracked),)

julia> Tracker.gradient(x -> logabsgamma(x)[2], rand())
(0.0 (tracked),)

[mcabbott]

Looks fine. Test failure on v1 is kron as in #125

[devmotion]

Bump 🙂

[mcabbott]

Ah I assumed you had merge rights here, sorry.

[devmotion]

Thank you!