Enable opting out of TSVI

[penelopeysm]

Performance gains in #1132 are only really significant when not using TSVI.

TSVI gets used more than it should (#1086).

This PR provides a mechanism to globally opt out of TSVI, in a way that is fully backwards-compatible.

Changing this to a global opt-in would just require changing the default to false.

A per-model opt-in would be doable too, albeit requiring more work. I think it's better to start from a global one and only consider a per-model one if people complain.

Performance

Edit: benchmarking this, I just realised that dereferencing USE_THREADSAFE_EVAL gives rise to more overhead than checking Threads.nthreads(). A per-model one should make this work better because if it's stored in the type, it can be removed at compile time. I'll investigate this further.

[github-actions]

Benchmark Report for Commit 6a0ecfa

Computer Information

Julia Version 1.11.7
Commit f2b3dbda30a (2025-09-08 12:10 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 │   Dim │  AD Backend │           VarInfo │ Linked │ t(eval)/t(ref) │ t(grad)/t(eval) │
├───────────────────────┼───────┼─────────────┼───────────────────┼────────┼────────────────┼─────────────────┤
│ Simple assume observe │     1 │ forwarddiff │             typed │  false │            7.5 │             1.7 │
│           Smorgasbord │   201 │ forwarddiff │             typed │  false │          757.0 │            43.8 │
│           Smorgasbord │   201 │ forwarddiff │ simple_namedtuple │   true │          436.1 │            55.2 │
│           Smorgasbord │   201 │ forwarddiff │           untyped │   true │          798.1 │            37.1 │
│           Smorgasbord │   201 │ forwarddiff │       simple_dict │   true │         7229.9 │            27.8 │
│           Smorgasbord │   201 │ forwarddiff │      typed_vector │   true │          776.1 │            42.4 │
│           Smorgasbord │   201 │ forwarddiff │    untyped_vector │   true │          795.6 │            37.9 │
│           Smorgasbord │   201 │ reversediff │             typed │   true │          920.4 │            45.5 │
│           Smorgasbord │   201 │    mooncake │             typed │   true │          734.1 │             6.0 │
│           Smorgasbord │   201 │      enzyme │             typed │   true │          919.1 │             4.0 │
│    Loop univariate 1k │  1000 │    mooncake │             typed │   true │         4035.9 │             5.7 │
│       Multivariate 1k │  1000 │    mooncake │             typed │   true │         1083.7 │             8.6 │
│   Loop univariate 10k │ 10000 │    mooncake │             typed │   true │        44505.8 │             5.4 │
│      Multivariate 10k │ 10000 │    mooncake │             typed │   true │         8929.4 │             9.9 │
│               Dynamic │    10 │    mooncake │             typed │   true │          130.1 │            13.0 │
│              Submodel │     1 │    mooncake │             typed │   true │            9.5 │            11.2 │
│                   LDA │    12 │ reversediff │             typed │   true │         1012.2 │             2.1 │
└───────────────────────┴───────┴─────────────┴───────────────────┴────────┴────────────────┴─────────────────┘

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 81.69%. Comparing base (62a8746) to head (6a0ecfa).

Additional details and impacted files

@@            Coverage Diff             @@
##           py/ldf    #1128      +/-   ##
==========================================
+ Coverage   81.67%   81.69%   +0.01%     
==========================================
  Files          42       42              
  Lines        3919     3922       +3     
==========================================
+ Hits         3201     3204       +3     
  Misses        718      718              

☔ 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.

[github-actions]

DynamicPPL.jl documentation for PR #1128 is available at:
https://TuringLang.github.io/DynamicPPL.jl/previews/PR1128/