Compile `SSMProblem` objects into `Distributions.ProductNamedTupleDistribution`

[yebai]

Useful to have a to_distribution(ssm_problem) which returns a Distributions.ProductNamedTupleDistribution. This would allow PPL users like JuliaBUGS and Turing to use SSMProblems inside @model easily.

TuringLang/Turing.jl#2659 and TuringLang/JuliaBUGS.jl#27 are related.

[THargreaves]

Might be misunderstanding what is desired here but I think this is superseded by the current Turing integration #67 / #134 which lets you convert a SSM model into a Distributions.jl distribution that you can sample from and compute log densities of.

This is just a regular multivariate distribution rather than a product distribution. I think the latter might not be appropriate as it is meant for a stack of independent distributions which is not the case for either the prior or filtered distribution of a SSM