`tovec` causing an error when sampling a Product named tuple distribution

[SamuelBrand1]

Following on from helpful chat in #929 I've found a bug in DPPL/Turing sampling from ProductNamedTupleDistribution around tovec being used down the stack.

Here is a reprex:

julia> using Distributions, Turing

julia> priors = (a = Normal(), b = Normal());

julia> priors_dist = product_distribution(priors)
# ProductNamedTupleDistribution{(:a, :b)}(
# a: Normal{Float64}(μ=0.0, σ=1.0)
# b: Normal{Float64}(μ=0.0, σ=1.0)
# )


julia> @model function sample_nt(priors_dist)
       x ~ priors_dist
       
       return x
       end
# sample_nt (generic function with 2 methods)

julia> mdl = sample_nt(priors_dist);

julia> mdl()

ERROR: MethodError: no method matching tovec(::@NamedTuple{a::Float64, b::Float64})
The function tovec exists, but no method is defined for this combination of argument types.

Closest candidates are:
tovec(::LinearAlgebra.Cholesky)
@ DynamicPPL ~/.julia/packages/DynamicPPL/vlXDM/src/utils.jl:446
tovec(::AbstractArray)
@ DynamicPPL ~/.julia/packages/DynamicPPL/vlXDM/src/utils.jl:445
tovec(::Real)
@ DynamicPPL ~/.julia/packages/DynamicPPL/vlXDM/src/utils.jl:444

[penelopeysm]

Yeah, ProductNamedTupleDistribution is fairly new and we don't have the machinery needed for it quite yet.

@sunxd3 would you be able to take a look at it? Might also be nice to add a demo model with it. If you don't have the time I can do it next week too.

[sunxd3]

Yeah sure, I'll take a look.

[penelopeysm]

So the initial MethodError is fixed, but more changes will be needed: TuringLang/Bijectors.jl#388