Different numeric types for the GammaShapeLikelihood constructor?

[abpolym]

In my new branch I am currently making some tests for the rules of the GammaMixture node, because there are none.

I encountered the following error:

MethodError: no method matching ReactiveMP.GammaShapeLikelihood(::Float32, ::Float64)
Closest candidates are:
  ReactiveMP.GammaShapeLikelihood(::T, ::T) where T<:Real

Do we want GammaShapeLikelihood to be also defined for different float numeric types ?

GammaShapeLikelihood(p::Real, γ::Real) = GammaShapeLikelihood(promote(p, γ)...)

We could also just enforce consistent floating type inside the rule but that sounds kind of sus:

@rule GammaMixture((:a, k), Marginalisation) (q_out::Any, q_switch::Any, q_b::GammaDistributionsFamily) = begin
    p = Float64(probvec(q_switch)[k])
    [...]

[bvdmitri]

I'm up to

GammaShapeLikelihood(p::Real, γ::Real) = GammaShapeLikelihood(promote(p, γ)...)