Add `basemeasure` method for discrete `Distribution`s

[cscherrer]

@cscherrer It seems some recent changes of Soss/MeasureTheory broke the tests: https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/pull/206/checks?check_run_id=3858536599#step:6:962 and https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/pull/206/checks?check_run_id=3858536599#step:6:1068. Tests passed on the master branch two weeks ago. Maybe it's related to cscherrer/Soss.jl#305?

Originally posted by @devmotion in JuliaGaussianProcesses/AbstractGPs.jl#206 (comment)

FWIW this passes:
https://gist.github.com/cscherrer/f2788b7ab62a232eb42a3068fe2fe56c

But it looks these tests are using Distributions instead.

The problem seems to be here:

julia> using MeasureTheory

julia> basemeasure(Poisson(3))
CountingMeasure(ℤ[0:ℵ₀])

julia> basemeasure(Dists.Poisson(3))
ERROR: MethodError: no method matching basemeasure(::Distributions.Poisson{Float64})
Closest candidates are:
  basemeasure(::Pullback{TransformVariables.CallableTransform{T}, Lebesgue{ℝ}}) where T<:TransformVariables.ScalarTransform at /home/chad/git/MeasureTheory.jl/src/combinators/transforms.jl:81
  basemeasure(::Pullback) at /home/chad/git/MeasureTheory.jl/src/combinators/transforms.jl:63
  basemeasure(::SuperpositionMeasure) at /home/chad/git/MeasureBase/src/combinators/superpose.jl:48
  ...
Stacktrace:
 [1] top-level scope
   @ REPL[3]:1

The solution is to add a basemeasure method for Distributions.Discrete so Poisson can use it, i.e.

julia> typeof(Dists.Poisson(3)) |> supertype
Distributions.Distribution{Distributions.Univariate, Distributions.Discrete}