Gibbs fixes for DPPL 0.37 (plus tiny bugfixes for ESS + HMC)

[penelopeysm]

CI status on this PR vs #2625 (pass / fail / error):

everything else #2625 - 1392 / 3 / 28 
everything else #2628 - 1421 / 2 / 10

Inference #2625 - 68 / 0 / 3
Inference #2628 - 68 / 0 / 3

Gibbs #2625 - 2248 / 33 / 26
Gibbs #2628 - 2511 / 4 / 10

I believe most (if not all) of the remaining test failures relate to pMCMC and VariableOrderAccumulator.

---

Well, let's see if I actually understand what I'm doing now. 🫠

In fairness to myself, before this PR, Gibbs would give wrong results like this:

julia> using Turing, Random

julia> @model function f()
           x ~ Normal()
           y ~ Normal(x)
           1.0 ~ Normal(y)
       end
f (generic function with 2 methods)

julia> chn = sample(Xoshiro(468), model, Gibbs(:x => HMC(0.1, 20), :y => MH()), 20000)
[...]
Summary Statistics
  parameters      mean       std      mcse     ess_bulk     ess_tail      rhat   ⋯
      Symbol   Float64   Float64   Float64      Float64      Float64   Float64   ⋯

           x   -0.0072    0.9965    0.0046   46727.7614   20309.9910    1.0000   ⋯
           y    0.6848    0.7669    0.0083    8493.5698    7860.8056    1.0003   ⋯
[...]

The correct result should be somewhere around (x, y) = (1/3, 2/3) (can be verified with NUTS).

With this PR:

julia> chn = sample(Xoshiro(468), model, Gibbs(:x => HMC(0.1, 20), :y => MH()), 20000)
[...]
Summary Statistics
  parameters      mean       std      mcse    ess_bulk    ess_tail      rhat   e ⋯
      Symbol   Float64   Float64   Float64     Float64     Float64   Float64     ⋯

           x    0.3416    0.8140    0.0084   9366.9879   6877.5666    1.0002     ⋯
           y    0.6854    0.8271    0.0167   2476.9080   1648.9534    1.0009     ⋯
[...]

On top of that, I also ran some ESS+ESS Gibbs tests on both main and this PR, and both branches gave the same results when seeded, which is quite reassuring.

[github-actions]

Turing.jl documentation for PR #2628 is available at:
https://TuringLang.github.io/Turing.jl/previews/PR2628/

[codecov]

Codecov Report

Attention: Patch coverage is 75.00000% with 2 lines in your changes missing coverage. Please review.

Project coverage is 24.57%. Comparing base (b0df6a6) to head (c0c298e).
Report is 1 commits behind head on mhauru/dppl-0.37-pmcmc.

Files with missing lines	Patch %	Lines
src/mcmc/ess.jl	0.00%	2 Missing ⚠️

Additional details and impacted files

@@                    Coverage Diff                     @@
##           mhauru/dppl-0.37-pmcmc    #2628      +/-   ##
==========================================================
+ Coverage                   24.47%   24.57%   +0.10%     
==========================================================
  Files                          22       22              
  Lines                        1479     1481       +2     
==========================================================
+ Hits                          362      364       +2     
  Misses                       1117     1117              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.