Formatting loaded values

[sethaxen]

Currently, we just return all loaded objects from JSON files without modification. JSON3 generates Objects, which are AbstractDict{Symbol}s but can also be accessed with string indices and a dot syntax. We may want to reformat these outputs for better compatibility with Julia. For example:

1. We want to make sure that loaded reference posterior draws implement the Tables interface. The eltype of the vector of draws should be the narrowest possible eltype. The idea is to make it straightforward for users to analyze the draws, e.g. by plotting them.
2. posteriordb stores matrices as lists of row vectors (https://github.com/stan-dev/posteriordb/blob/master/doc/DATABASE_CONTENT.md#datadata; because JSON has no way to encode matrices). To avoid confusion and allow use, we should package these as matrices.

Since posteriordb is very Stan-focused (only contains one pymc model) and will for now likely be used in conjunction with the StanJulia packages, a useful check would be that we can use all model code and data directly in StanSample.jl.

[sethaxen]

I propose we format all JSON outputs to be AbstractDict{String}s, as these still implement the Tables interface. We may also want to use OrderedCollections.OrderedDict to store the outputs, as this preserves the ordering in case that is important, and OrderedCollections is a light and common dependency.

julia> using Tables, OrderedCollections, DataFrames

julia> d = OrderedDict("x" => randn(100), "y" => randn(100));

julia> DataFrame(columntable(d))
100×2 DataFrame
 Row │ x           y         
     │ Float64     Float64   
─────┼───────────────────────
   1 │  0.194559    1.88364
   2 │  0.514289   -0.979216
   3 │  0.731907   -0.262208
   4 │ -0.232648    1.93591
  ⋮  │     ⋮           ⋮
  98 │  0.613435    0.404689
  99 │  0.541503   -0.654151
 100 │  0.0387091  -0.261594
              93 rows omitted