Chains internals shouldn't use (undocumented) reserved names

[penelopeysm]

julia> using Turing

julia> @model f() = lp ~ Normal()
f (generic function with 2 methods)

julia> sample(f(), NUTS(), 100)
┌ Info: Found initial step size
└   ϵ = 1.6125
Sampling 100%|████████████████████████████████████████████████████████████████████| Time: 0:00:03
Chains MCMC chain (100×13×1 Array{Float64, 3}):

Iterations        = 51:1:150
Number of chains  = 1
Samples per chain = 100
Wall duration     = 4.45 seconds
Compute duration  = 4.45 seconds
parameters        =
internals         = lp, n_steps, is_accept, acceptance_rate, log_density, hamiltonian_energy, hamiltonian_energy_error, max_hamiltonian_energy_error, tree_depth, numerical_error, step_size, nom_step_size

Summary Statistics
  parameters   mean   std      mcse   ess_bulk   ess_tail      rhat   ess_per_sec
      Symbol    Any   Any   Float64    Float64    Float64   Float64       Float64


Quantiles
  parameters   2.5%   25.0%   50.0%   75.0%   97.5%
      Symbol    Any     Any     Any     Any     Any

We can't use lp as a parameter name because it's one of the internal symbols.

I didn't check the other internals but I assume similar errors would occur.

I suppose that here lp is being provided to MCMCChains by an upstream caller (something in Turing itself probably), but in any case internals + parameters should be kept in strictly separate spaces - a parameter lp and an internal lp have different meanings.

Edit: Turns out that chn[:lp] still gives you the parameter values, so the problem is technically that the logprobs are no longer accessible, plus the output display is wrong, plus the grouping of lp into internals is wrong.