fix(kernel): boolean volume corruption in coaxial cylinder/annulus stacks

[ecto]

Regression from a turbopump field repro: intersecting two deep union
trees of coaxial cylinders/annuli returned ~661 mm³ of phantom volume
with only ~100 mm² of surface (isoperimetrically impossible), while
every pairwise intersection of the same parts was correct. The new
catalogue-style test (torr_phantom_intersection.rs) asserts the
intersection is empty and pins both union trees to their analytic
volumes; those controls exposed four kernel bugs, all fixed:

• face_sample_point picked classification samples ON the outer boundary
  of planar faces with holes — exactly where the other solid's surface
  passes, since split curves come from it — so ray casts against the
  other operand's tessellation misclassified split sub-faces (a pocket
  cut over a coaxial bore read +48% volume). Samples now maximize
  clearance from every boundary (outer, polygon holes, degenerate
  circle holes) and validate with point_in_face.
• Disk sub-faces created by circle splits inherited the SSI circle's
  arbitrary winding instead of the loop-CCW-around-surface-normal
  convention the tessellator expects; a surviving disk (bore mouth kept
  as the visible floor under a press-fit ring) tessellated inside-out
  and integrated negative volume.
• A boss ring exactly inscribed to a cap rim (tangent from inside)
  never punched its hole through the cap: the containment test required
  strict clearance, and a circle equal to the cap's own rim is now
  rejected explicitly instead (degenerate-loop analog of
  circle_is_own_boundary).
• Circles partially overhanging a degenerate cap's rim could not split
  it at all: the degenerate path now polygonizes the rim and reuses the
  arc splitter, split_planar_face_by_arc re-homes pre-existing inner
  loops onto the containing sub-face instead of dropping them, and the
  pipeline's split gate probes 8 rim angles instead of a single fixed
  point that could land outside the face.

Full static union tree error: +4.7% → +0.07%; rotating tree +0.9% →
+0.3%. cargo test --workspace green, clippy -D warnings clean.

Co-Authored-By: Claude Fable 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01JvTnDfeNNvXh1E89t387PW

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

4 Skipped Deployments

Project	Deployment	Actions	Updated (UTC)
mecheval	Ignored		Jul 7, 2026 4:36pm
vcad	Ignored		Jul 7, 2026 4:36pm
vcad-docs	Ignored		Jul 7, 2026 4:36pm
vcad-mcp	Ignored		Jul 7, 2026 4:36pm

[chojiai]

What shipped

Intersecting complex assemblies of coaxial cylinders and rings — such as a shaft floating inside stacked bearing bores — could produce phantom solid volume that was geometrically impossible: the reported material was far too large to fit inside the reported surface area. Four related bugs in the geometry engine caused this, and all four are now fixed.

• Boolean operations on nested coaxial assemblies now return correct results; the specific field case (a turbopump rotating group intersected with its static housing) went from reporting ~661 mm³ of phantom material to correctly reporting an empty intersection.
• Inspection and integrity reports now include surface area alongside volume, and will warn explicitly when a part's volume and surface area are mathematically inconsistent — a sign the volume calculation cannot be trusted.

---

_{Plain-English summary generated by Choji from this pull request.}