Remove bogus AbstractFloat conversion.

[tpapp]

Fixes

julia>using ForwardDiff: Dual
julia>AbstractFloat(Dual(1.0, 2.0)) isa AbstractFloat
false

[tpapp]

FWIW, I think the way forward for the host of stack overflow issues (JuliaMath/SpecialFunctions.jl#186, #324, etc) is to remove the float(::Dual) method as well.

Dual is not <: AbstractFloat, so that method violates its contract.

[KristofferC]

FWIW, was added in 2e8d016.

[tpapp]

Friendly bump: I am happy to work on this if anything else is required.

[sethaxen]

Another friendly bump. What needs to be done to get this PR merged?

[nalimilan]

FWIW, I think the way forward for the host of stack overflow issues (JuliaMath/SpecialFunctions.jl#186, #324, etc) is to remove the float(::Dual) method as well.

Dual is not <: AbstractFloat, so that method violates its contract.

Note that float(::Complex) also returns a Complex, so the contract of that function isn't to return an AbstractFloat. I can't comment on what's the best choice for Dual, though.

[andreasnoack]

See JuliaMath/SpecialFunctions.jl#186 (comment). The issue with loggamma was fixed a while ago but it keeps showing up every time SpecialFunctions make a new breaking release because older versions of DiffRules didn't have appropriate compat bounds.