(HalfEdgeMesh) Improve interface for iteration

[csparker247]

Summary

Two related problems with the current iteration interface:

1. All iteration methods return std::vector by value (P1)
vertices(), edges(), faces(), vertices_interior(), vertices_boundary(), wheel(), boundaries(), outgoing_edges(), and connected_components() all allocate and return a new std::vector on every call. The ABF/ABFPlusPlus solvers call these in tight per-iteration loops, causing O(n) heap allocations per solver step.

2. FacePtr iteration requires *face dereference (A2)
Range-based iteration over a face pointer requires for (const auto& e : *face) rather than for (const auto& e : face->edges()).

Proposed Fix

Replace vector-returning methods with lazy range views:

• std::span (or similar) for contiguous storage (vertices, edges, faces)
• Custom iterator/sentinel for linked-list traversals (wheel, outgoing_edges, boundaries, face edges)

Add face->edges() as a named iterable accessor.

Acceptance Criteria

• Hot-path iteration does not heap-allocate on every call
• face->edges() works in range-based for loops
• for (const auto& e : *face) continues to work (no breaking change)
• All existing tests pass

Conductor Tracks

• P1 — iteration redesign
• A2 — face->edges() accessor (subsumed by P1)

---

Migrated from GitLab issue #3