Parallelization of loo using mirai and mori

[florence-bockting]

Summary

Fixes #308

• Replaces parallel::mclapply() / parLapply() with mirai + mori for per-observation parallelism (cross-platform, including Windows).
• Introduces two recommended parallelism workflows:
  • User-managed pools via mirai::daemons() (local, remote/SSH/HPC); a connected pool always wins.
  • Map-style jobs via exported loo_mirai() (optional n_daemons for one-off pools).
• Deprecates the cores argument and options(mc.cores) as loo parallelism controls (still functional as a per-call local pool bridge when > 1 and no pool is connected; once-per-session warning).
• Removes options(loo.cores) (deprecated since 2.x).

What changed

Core (R/parallel.R): with_loo_daemons(), loo_map(), loo_pool_is_local(), loo_n_workers(), loo_mirai(); deprecation helpers for cores / mc.cores.

Parallelized functions: loo() (function method), psis()/sis()/tis(), relative_eff(), loo_subsample(), loo_moment_match(), loo_model_weights().

Pool precedence: any connected mirai::daemons() pool (user-managed) always wins > deprecated cores is ignored. Local pools use mori zero-copy for broadcast objects (e.g. draws); remote pools serialize (with conservative chunking when broadcasting).

Also: mirai + mori in DESCRIPTION; vignettes/loo2-parallel.Rmd; vignettes/migration-guide.Rmd;
tests/testthat/test_parallel.R; benchmark/ scripts + bench-comparison.md.

Review guide

1. vignettes/loo2-parallel.Rmd shows the user-facing model
2. vignettes/migration-guide.Rmd provides a parallelization migration table and examples
3. R/parallel.R includes the pool lifecycle + loo_map() transport, loo_mirai()
4. tests/testthat/test_parallel.R include serial/parallel equivalence, pool precedence, loo_mirai(), deprecation warnings
5. benchmark/README.md includes first attempt of a small baseline vs new comparison (see first results in benchmark/bench-comparison.md)

Initial benchmarks (Linux, one machine): loo.function + large draws benefits most (~4× with persistent pool); matrix psis() does not (communication-bound); per-call pool pays ~1s spawn/teardown per call.

Follow-up work

• Reviewing: Please have a look at the current implementation and check it for correctness and usability. Any comments and improvements are welcome.
• Benchmarking: master vs this branch across problem sizes, all (or selected number of) parallelized functions, OSes (Linux/macOS/Windows), and metrics (wall-clock, allocation, peak RSS). You can for example extend benchmark/.
• LSAT case study: posteriorDB lsat-data; showcase speedup and all three parallelism modes (one-off, persistent pool, simulation loop).
• Remote SSH: two-machine test; verify correctness, measure speedup, document setup (mirai::daemons(url = ..., remote = ssh_config(...))).
• Documentation: expand vignettes/loo2-parallel.Rmd with assumptions, when-to-use guidance, function-specific notes, memory model.

Current limitations of implementation

• Matrix psis() rarely speeds up (large data shipped per worker).
• Per-call cores > 1 can be slower than serial on small problems without loo.daemons.
• Remote SSH untested in CI

This PR has been developed with assistance of Cursor.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 30834a4 is merged into loo-v3.0.0:

• ❗🐌loo_function: 1.91s -> 2.03s [+6%, +6.91%]
• ❗🐌loo_matrix: 1.71s -> 1.75s [+1.73%, +2.75%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 97.65625% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 94.39%. Comparing base (fe16a43) to head (30834a4).

Files with missing lines	Patch %	Lines
R/parallel.R	95.72%	5 Missing ⚠️
R/psis_approximate_posterior.R	50.00%	1 Missing ⚠️

Additional details and impacted files

@@              Coverage Diff               @@
##           loo-v3.0.0     #378      +/-   ##
==============================================
+ Coverage       92.70%   94.39%   +1.68%     
==============================================
  Files              31       32       +1     
  Lines            3029     3121      +92     
==============================================
+ Hits             2808     2946     +138     
+ Misses            221      175      -46     

☔ View full report in Codecov by Harness.
📢 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.