Remove VariableOrderAccumulator and subset and merge on accs

[mhauru]

These are no longer needed in Turing.jl.

To be merged after #901. Merging that will also clean up the diff of this. Please don't try to review before that.

[github-actions]

Benchmark Report for Commit 1d93529

Computer Information

Julia Version 1.11.6
Commit 9615af0f269 (2025-07-09 12:58 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 4 × AMD EPYC 7763 64-Core Processor
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, znver3)
Threads: 1 default, 0 interactive, 1 GC (on 4 virtual cores)

Benchmark Results

|                 Model | Dimension |  AD Backend |      VarInfo Type | Linked | Eval Time / Ref Time | AD Time / Eval Time |
|-----------------------|-----------|-------------|-------------------|--------|----------------------|---------------------|
| Simple assume observe |         1 | forwarddiff |             typed |  false |                  8.3 |                 1.5 |
|           Smorgasbord |       201 | forwarddiff |             typed |  false |                622.9 |                44.2 |
|           Smorgasbord |       201 | forwarddiff | simple_namedtuple |   true |                297.7 |                73.5 |
|           Smorgasbord |       201 | forwarddiff |           untyped |   true |               1166.3 |                28.5 |
|           Smorgasbord |       201 | forwarddiff |       simple_dict |   true |               6381.5 |                28.7 |
|           Smorgasbord |       201 | reversediff |             typed |   true |                992.5 |                41.0 |
|           Smorgasbord |       201 |    mooncake |             typed |   true |                962.0 |                 4.3 |
|    Loop univariate 1k |      1000 |    mooncake |             typed |   true |               5487.4 |                 4.0 |
|       Multivariate 1k |      1000 |    mooncake |             typed |   true |                952.2 |                 9.1 |
|   Loop univariate 10k |     10000 |    mooncake |             typed |   true |              62501.9 |                 3.5 |
|      Multivariate 10k |     10000 |    mooncake |             typed |   true |               8137.0 |                10.0 |
|               Dynamic |        10 |    mooncake |             typed |   true |                126.3 |                12.3 |
|              Submodel |         1 |    mooncake |             typed |   true |                 12.3 |                 5.4 |
|                   LDA |        12 | reversediff |             typed |   true |               1268.0 |                 3.3 |

[github-actions]

DynamicPPL.jl documentation for PR #1005 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1005/

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.16%. Comparing base (f0ac109) to head (1d93529).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1005      +/-   ##
==========================================
+ Coverage   81.91%   82.16%   +0.24%     
==========================================
  Files          38       38              
  Lines        4025     3935      -90     
==========================================
- Hits         3297     3233      -64     
+ Misses        728      702      -26     

☔ 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.

[coveralls]

Pull Request Test Coverage Report for Build 16680558902

Details

• 621 of 739 (84.03%) changed or added relevant lines in 27 files are covered.
• 62 unchanged lines in 11 files lost coverage.
• Overall coverage decreased (-0.8%) to 82.369%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/utils.jl	8	9	88.89%
src/context_implementations.jl	31	33	93.94%
src/experimental.jl	0	2	0.0%
src/pointwise_logdensities.jl	49	51	96.08%
src/model.jl	33	36	91.67%
ext/DynamicPPLJETExt.jl	0	4	0.0%
src/accumulators.jl	28	32	87.5%
src/threadsafe.jl	36	40	90.0%
src/values_as_in_model.jl	20	24	83.33%
src/default_accumulators.jl	37	42	88.1%

Files with Coverage Reduction	New Missed Lines	%
src/context_implementations.jl	1	93.33%
src/debug_utils.jl	1	88.56%
src/test_utils/contexts.jl	1	88.89%
src/contexts.jl	2	74.45%
src/logdensityfunction.jl	3	57.14%
src/utils.jl	5	71.28%
src/distribution_wrappers.jl	6	28.57%
src/simple_varinfo.jl	7	70.86%
src/abstract_varinfo.jl	9	76.76%
src/threadsafe.jl	9	65.25%

Totals
Change from base Build 16174747412:	-0.8%
Covered Lines:	3233
Relevant Lines:	3925

---

💛 - Coveralls

[penelopeysm]

As @yebai requested #901 (comment) I'll add some extra elaboration in the changelog about why we no longer need num_produce / order / VariableOrderAccumulator. Leaving this comment here as a reminder to myself to do this next week.

(I'm happy to do this since the corresponding Turing PR was mine.)

Done.

[penelopeysm]

num_produce / order / pMCMC section in changelog has been expanded.

[penelopeysm]

There's a set_retained_vns_del! in src/threadsafe.jl that should also be removed, but otherwise very happy

[penelopeysm]

Probably that method never needed to exist because in Turing's particle MCMC code threadsafe evaluation is always disabled:

https://github.com/TuringLang/Turing.jl/blob/d75e6f269b18800e366b383b5059ed08e10ad7a7/src/mcmc/particle_mcmc.jl#L386-L390

I actually wonder whether threadsafe execution would work now with the changes we have... my guess is not but idk.