test(boolean): un-ignore recovered boolean tests

[andymai]

A full re-run of every ignored boolean test showed several now pass reliably — their ignore attributes were stale after recent determinism and same-domain fixes (#748 et al).

Changes

• Un-ignores 9 tests across boolean_stress, boolean_edge_cases, boolean_invariants, coincident_proptest, and the lib boolean suite. Formerly-flaky compound-cut tests were gated on 10/10 passes in fresh processes before un-ignoring.
• compound_cut_matches_sequential_3x3_grid: corrects its volume oracle (box_vol was 15·15·2=450 against an actual 10×10×2 box, so the < box*0.99 assertions could never fail) and keeps it ignored — with the honest oracle it flakes ~2/3 of runs (compound_cut multi-tool path produces non-manifold, non-deterministic, incorrect results #747).
• Left ignored: compound_cut_matches_sequential_4x4_grid (borderline under-cut, intermittent) and staircase_fuse_with_cylinders (~125s, CI-time choice).

Verification

fmt, clippy -D warnings, boundary check, full operations lib + 4 integration targets green.

[github-actions]

WASM Binary Size

File	main	PR	Delta
brepkit_wasm.wasm	4.70 MB	4.70 MB	0 KB (0%)

✅ No size change.

[Copilot]

Pull request overview

This PR refreshes the boolean test suite by re-enabling previously-ignored tests that now pass reliably after recent determinism/same-domain fixes, and by correcting a stale volume oracle in a still-ignored flaky compound-cut test.

Changes:

• Un-ignored 9 boolean-related tests across stress, edge-case, invariant, coincident-face proptests, and the boolean module’s internal test suite.
• Corrected compound_cut_matches_sequential_3x3_grid’s box-volume oracle (while keeping the test ignored due to ongoing flakiness tracked in #747).
• Preserved existing ignores for known-slow or still-flaky tests (e.g., 4×4 grid, staircase benchmark).

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
crates/operations/tests/coincident_proptest.rs	Re-enables Z-rotation invariance proptest for coincident face-stack fusing.
crates/operations/tests/boolean_stress.rs	Re-enables thin-wall cut and asymmetric fuse stress tests.
crates/operations/tests/boolean_invariants.rs	Re-enables a cylinder-from-box cut volume invariant test.
crates/operations/tests/boolean_edge_cases.rs	Re-enables sequential vertex-drift and alternating fuse/cut edge-case tests.
crates/operations/src/boolean/tests.rs	Re-enables several boolean regression/compound-cut tests; corrects the 3×3 grid volume oracle while keeping it ignored.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[greptile-apps]

Greptile Summary

This PR un-ignores 9 boolean tests that were previously skipped due to GFA pipeline limitations or flakiness now resolved by earlier determinism fixes (#748), and corrects the volume oracle in the still-ignored 3\u00d73 compound-cut test (was 15\u00d715\u00d72=450, should be 10\u00d710\u00d72=200).

• 9 tests un-ignored across boolean_stress, boolean_edge_cases, boolean_invariants, coincident_proptest, and the lib boolean suite \u2014 each verified as 10/10 passes in fresh processes before promotion.
• 3\u00d73 oracle fix: box_vol was wrong by 2.25\u00d7, meaning the < box*0.99 assertion could never have failed even on a completely broken result; corrected and the test remains ignored per compound_cut multi-tool path produces non-manifold, non-deterministic, incorrect results #747.
• Kept ignored: compound_cut_matches_sequential_4x4_grid (intermittent) and staircase_fuse_with_cylinders (~125 s runtime).

Confidence Score: 5/5

Safe to merge — all changes are test-only, removing stale ignore attributes and fixing an incorrect oracle constant.

No production logic is touched. The oracle fix for the 3×3 grid is correct, and the two nits are documentation/hardening concerns that do not affect correctness.

The 4×4 grid ignore message in crates/operations/src/boolean/tests.rs still carries a stale claim fixed for 3×3 but not updated for 4×4.

Important Files Changed

Filename	Overview
crates/operations/src/boolean/tests.rs	Corrects the 3×3 grid oracle, un-ignores three tests; the 4×4 ignore message still has a stale 'oracle passes on garbage' claim.
crates/operations/tests/coincident_proptest.rs	Un-ignores prop_face_stack_z_rotation_invariant; 16-case proptest budget may not reliably hit the historically bad 3.98 rad angle.
crates/operations/tests/boolean_edge_cases.rs	Clean un-ignores of test_sequential_boolean_vertex_drift and test_alternating_union_cut; no logic changes.
crates/operations/tests/boolean_invariants.rs	Clean un-ignore of cut_cylinder_from_box_volume (formerly GFA limitation); test logic unchanged.
crates/operations/tests/boolean_stress.rs	Clean un-ignores of thin_wall_cut and fuse_asymmetric_boxes; no test logic changes.

Prompt To Fix All With AI

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
crates/operations/src/boolean/tests.rs:1721-1725
The `#[ignore]` reason on the 4×4 grid still says "the `` `< box*0.99` oracle passes on garbage``," but that bug only existed in the 3×3 grid — the 4×4 oracle was already correct (`20.0 * 20.0 * 2.0 = 800` matches `make_box(20, 20, 2)`). This was fixed for the 3×3 test in this PR; the 4×4 message was not updated and will mislead anyone reading it when evaluating whether to un-ignore the test.

```suggestion
#[ignore = "flaky — multi-tool compound/sequential cuts through the mesh-boolean \
            fallback are non-deterministic across processes (seed-dependent vertex \
            welding) and under-cut faceted re-input; borderline under-cut intermittent. \
            Tracked in #747; revisit under the GFA rewrite."]
fn compound_cut_matches_sequential_4x4_grid() {
```

### Issue 2 of 2
crates/operations/tests/coincident_proptest.rs:57-72
**Proptest case budget may not cover the historically bad angle**

The whole proptest block runs `with_cases(16)`, so `prop_face_stack_z_rotation_invariant` gets 16 draws from `[0, 2π)`. The old ignore described up to ~15% volume drift specifically at angles like 3.98 rad. With 16 uniform samples across a range of 6.28, the probability of landing within ±0.05 of that angle is ≈ 25% per run — meaning ~75% of CI runs will never exercise the historically worst input. A dedicated deterministic regression case at `angle = 3.98_f64` alongside the proptest would give permanent assurance that the specific failure is gone.

_{Reviews (2): Last reviewed commit: "test(boolean): correct tool description ..." | Re-trigger Greptile}