Support ProductNamedTupleDistribution, export output_size

[penelopeysm]

Closes #388.

• Implement for simple cases
• Make it type stable
• Handle nested ProductNamedTupleDistribution
• Implement logabsdetjac

julia> using Bijectors

julia> d = product_distribution((a=Normal(), b=LogNormal()))
ProductNamedTupleDistribution{(:a, :b)}(
a: Normal{Float64}(μ=0.0, σ=1.0)
b: LogNormal{Float64}(μ=0.0, σ=1.0)
)


julia> b = bijector(d)
Bijectors.NamedStacked{(:a, :b), @NamedTuple{a::typeof(identity), b::Base.Fix1{typeof(broadcast), typeof(log)}}, @NamedTuple{a::Int64, b::Int64}}((a = identity, b = Base.Fix1{typeof(broadcast), typeof(log)}(broadcast, log)), (a = 1, b = 2))

julia> x = rand(d)
(a = -0.2857657609237111, b = 1.7017383063757614)

julia> y = transform(b, x)
2-element Vector{Float64}:
 -0.2857657609237111
  0.5316502617934541

julia> x2 = transform(binv, y)
(a = -0.2857657609237111, b = 1.7017383063757614)

[github-actions]

Bijectors.jl documentation for PR #418 is available at:
https://TuringLang.github.io/Bijectors.jl/previews/PR418/

[sunxd3]

The code looks good to me.

One concern is that transformed(d) won't work at the moment because it still uses Base.size but I don't mind push this later if keen to merge.

[penelopeysm]

Ooh, okay. I am not sure how to implement that, let me look at it

[penelopeysm]

Thanks @sunxd3!