test/ci: repair compliance suite; demote weekly-compliance to placeholder until 0.2.0 on CRAN

[MaartenMarsman]

The weekly-compliance job had been red since April. Investigating it surfaced a chain of issues — fixed the real ones, and demoted the CI job to a placeholder because bitwise-in-CI is fundamentally unworkable until there's a matching CRAN release.

Genuine fixes (kept)

• CI infra: the job died before running anything — remotes missing, a 404ing/unused mixedGM install, and library(bgms) with the package never installed. (Moot now that the job is a placeholder, but the diagnosis drove the rest.)
• Stale output-property names: the harness read fit$posterior_coclustering_matrix (renamed → posterior_mean_coclustering_matrix in 0.2.0) and fit$posterior_mean_indicator for bgmCompare (gone; bgmCompare has only posterior_summary_indicator). Updated to the current API. (Confirmed these were renames, not removals, and not from this session's refactors.)
• Config drift → single source of truth: generate_fixtures.R and test_compliance.R duplicated the config matrix in different formats and had diverged (test side pinned the CRAN-era target_accept=0.6/0.65; generator used the default). Extracted a shared tests/compliance/config_defs.R both source, with target_accept pinned explicitly per config (NUTS 0.80, adaptive-metropolis 0.44).
• Re-baselined fixtures to the current 0.2.0 build (bitwise-vs-CRAN-0.1.6.3 is no longer meaningful).

Why the CI job is now a placeholder

Bitwise compliance against committed fixtures can't run in CI: NUTS leapfrog trajectories are chaotically FP-sensitive, so fixtures reproduce bitwise only on the machine that generated them — verified: ubuntu CI 0/32, macOS CI 8/32, vs 32/32 on the generating machine. So the bitwise check is inherently local (test_compliance.R, documented LOCAL-ONLY).

weekly-compliance.yaml is reduced to a green placeholder that documents the planned replacement: once 0.2.0 is on CRAN, install the CRAN release + dev build on one runner and diff them bitwise on the fly — same FP environment, so differences are real code changes; platform-independent; no committed fixtures.

Not in scope

• Continuous GGM / mixed-MRF configs (suite is discrete-only; the CRAN baseline predated them) — add when the CRAN-vs-dev job lands.
• The nightly-validation failures (2 Rhat-threshold tests + ~964 deprecation warnings) are a separate S9 thread.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.74%. Comparing base (bab7ca2) to head (22f6546).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #143      +/-   ##
==========================================
- Coverage   87.75%   87.74%   -0.01%     
==========================================
  Files          87       87              
  Lines       12867    12862       -5     
==========================================
- Hits        11291    11286       -5     
  Misses       1576     1576              

☔ 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.