Add narrow cylinder lowering and boundary PointOnFixedEdge recovery

[lucia3e8]

Summary

• Narrow cylinder theta-z lowering: for cylindrical faces with angular span < 30° (π/6), use (θ, z) coordinates instead of the default XY projection. The default projection collapses narrow cylinders to near-zero width, causing CDT failures.
• Boundary PointOnFixedEdge recovery: when the CDT reports a boundary point lying on a constrained edge, find that edge and split it at the point, then retry. Previously only Steiner points had recovery (reassign to pts[0]); boundary points caused immediate failure.
  • Uses a relaxed epsilon fallback (1000× snap_eps) to catch nearly-collinear cases that the initial search misses

These fixes address the PointOnFixedEdge and HalfEdgeInvariant errors that appeared on component FH12-22S-0.5SH_55 after the previous CDT preprocessing changes.

Batch test: 90/92 maintained (no regression, prepares for next fix)

Test plan

• cargo test passes
• Batch test: 90/92 pass (no regressions from PR Add DegenerateToroidalSurface support and CDT edge preprocessing #1)
• Narrow cylinder faces in XT30PW-F render correctly
• FH12-22S-0.5SH_55 boundary PointOnFixedEdge errors are caught and retried

🤖 Generated with Claude Code

Stack: 2/4 — prev: fix/degenerate-toroidal-surface, next: fix/collinear-perturbation

---

Note

Medium Risk
Touches core surface lowering and CDT error-recovery logic; while changes are targeted, they can affect mesh topology/quality and retry behavior on tricky geometry.

Overview
Adds a narrow-cylinder path for Surface::Cylinder that detects small angular spans during prepare() and lowers points using (theta, z) instead of the existing scaled-XY projection to avoid near-collinear 2D inputs.

Extends CDT retry handling in advanced_face so PointOnFixedEdge on boundary points triggers constrained-edge splitting (with a scale-relative snap_eps and a relaxed fallback), rather than failing immediately; Steiner-point recovery remains as point deduping.

^{Written by Cursor Bugbot for commit 4918813. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

[cursor]

Angular span calculation ignores atan2 wraparound at ±π

Medium Severity

The angular_span calculation uses angle_max - angle_min from raw atan2 values, which doesn't account for the discontinuity at ±π. atan2 returns values in (-π, π], so a narrow face positioned diametrically opposite the cylinder's ref_direction (near angle ±π) would have some vertices with angles just below π and others just above -π. This yields angular_span ≈ 2π instead of the true small span, causing narrow to be false when it needs to be true. The face then uses the degenerate scaled-XY projection — exactly the failure this PR aims to fix.