Research: ellipse and tangent path-repeat geometry (#110)

[madawei2699]

Research/prototype for #110

This is a bounded representation study, not a production geometry change. It adds no public ComponentPlan types or compiler behavior.

Findings

• Current CircleRing / RadialRepeat cover circular shells and circular tangent-oriented repeats, but not ellipse perimeter geometry or partial-path repetition.
• An oval shell workaround compiles as two circular caps plus two straight tangent walls: 4 components / 938 JSON bytes / 2,240 blocks, but the result is a stadium rather than an ellipse.
• An oval two-level arcade workaround requires 48 manually positioned Instances (plus a fixed-X assembly): 6,065 JSON bytes / 768 blocks. It is valid but bays do not rotate to local tangent.
• Recommendation: proceed with a focused EllipseRing + ellipse-only PathRepeat implementation follow-up. Do not build generic CurvePath or CAD/spline semantics.

Prototype evidence

• Elliptical shell: 2,412 unique voxels.
• Two-level tangent-oriented elliptical colonnade: 816 unique voxels.
• Open ellipse arc endpoint probe: count 9 includes both endpoints, 25 unique voxels.
• Deterministic SVG artifacts are included in prototypes/issue-110/artifacts/.

The spike is intentionally outside core exports/schema. It uses integer bounds, deterministic nearest-voxel tangent rotation, validation, and explicit block budgets. The SVGs are geometry illustrations, not MinePilot's six-view preview captures; the report calls out remaining integration checks for support diagnostics, bounds estimates, and in-game voxel quality.

See prototypes/issue-110/RESEARCH.md for the capability table, option comparison, API sketch, semantics, and limitations.

Validation

• Core build passed.
• Both current-vocabulary ComponentPlans compiled successfully.
• Three prototype fixtures ran; repeated fixture generation produced identical SVG hashes.
• Push hook also passed lint and the repository test suite (199 tests).

No production feature is being proposed for merge in this PR; keep this PR Draft until the research decision is reviewed.

[madawei2699]

Research decision accepted and carried forward to #117. This PR is intentionally prototype/research evidence only; its SVG/prototype code should not become production CraftDAG surface. Closing without merge while preserving the research history here.