Remove passing large amounts of NamedTuples in return value of Turing models

[SamuelBrand1]

At the moment, we are passing a lot of NamedTuples with "auxiliary" variables as generated quantities / return object from constructed Turing models.

This is mainly used internally via @submodel macro usage on constructing Turing models that represent various components of the epidemiological model we want to model generatively.

I don't think passing all these variables is necessary, cause code bloat and could be making type inference more difficult (I'm not so sure about this though). I think we can eliminate to only pass samples of the latent models.