Evaluate generic ellipse and tangent path-repeat geometry #110

Evidence

MinePilot issue #73 stress-tested three reusable families in an isolated worktree: a CircleRing/RadialRepeat circular arena control, a manually approximated elliptical shell, and a three-level elliptical arcade using ArcadeRun plus cap arcs. The full local report is in i365dev/MinePilot issue #73 worktree under docs/ISSUE_73_CURVED_RADIAL_STRESS_TEST.md.

Findings

• CircleRing is sufficient for a basic circular shell, but endpoint samples can trigger disconnected/floating support diagnostics.
• The elliptical shell requires four partial CircleRing arcs with separate centers/radii plus hand-authored tangent walls; it is visibly faceted and brittle to repair.
• The multi-level arcade can use ArcadeRun on straight tangent sides, but bays cannot follow an oval path or rotate to local tangent orientation. Cap arcs and broad platforms obscure the repeated bays.
• Both the shell and arcade families independently reproduce the same generic gap: no bounded ellipse/curve path and no path-based tangent-oriented repeat. This is not a request for a Colosseum-specific primitive.

Proposed evaluation

Please evaluate a small orthogonal extension, in sequence:

1. An EllipseRing or bounded CurvePath that deterministically expands to existing voxel nodes.
2. A PathRepeat primitive/authoring helper for assemblies with explicit, quantized rotation semantics.
3. Endpoint support diagnostics and safe prototype anchoring for CircleRing + RadialRepeat.

Require fixtures for both an elliptical shell and a multi-level curved arcade before accepting an implementation. Keep true freeform meshes, smooth curves, and landmark-specific components out of scope. Related: #83 and #105.