mesh: unify decimate_ring onto the public polyline simplifier

[JArmandoAnaya]

Filed by #244 (p7-s4), which added the public simplifier and deliberately did not do this.

The duplication

p7-s4 adds roadmaker::simplify_polyline in core/include/roadmaker/geometry/simplify.hpp — Ramer–Douglas–Peucker over an open polyline, returning the indices kept, iterative rather than recursive.

core/src/mesh/surface_boundary.cpp has had a private static decimate_ring since p2-s7. It solves the same problem for closed rings, via Clipper2's RamerDouglasPeucker, returning points.

Two RDP implementations in one kernel is one too many.

Why p7-s4 did not unify them

Not oversight — deliberate, and stated in the plan. decimate_ring gates every interior Steiner point in all three surface fill pipelines. Changing which vertices it keeps moves geometry in every surface, which breaks the byte-identity assertions that a large part of the mesh suite rests on. That is precisely the hazard #442 documented when it fixed Clipper2Lib::PointInPolygon, and it does not belong inside a feature sprint for an unrelated subsystem.

Scope

• Extend simplify_polyline to closed rings (or add a simplify_ring beside it) and re-express decimate_ring on top.
• Measure before and after: how many surfaces change, and by how much. The mesh: fill_backend::inside_region uses Clipper2 PointInPolygon on double paths, which is unsound below 1 m #442 method is the one to copy — a runtime control, a per-geometry diff, and an honest statement of which cases moved and whether they improved. mesh: fill_backend::inside_region uses Clipper2 PointInPolygon on double paths, which is unsound below 1 m #442 found that "every changed case improves" was false, and had to amend its own commit message; assume the same here.
• Regenerate whatever golden data legitimately moves, in the same PR, with the count in the message.
• Delete the Clipper2 round-trip if it turns out to be the only reason surface_boundary.cpp converts to PathD at that point.

Do not

Fold this into an unrelated sprint. It is a mesh-output change wearing a refactor's clothes.