Risk H1: measure the 3D packing density numbers in C# (harness --pack3d rows)

[libishm1]

From the risk register (wiki/research/RISK_REGISTER.md, H1): every 3D packing density figure quoted in the docs (33.8/35.7/38.7/53.3/49.3%) is a Python pybullet+VHACD or Rhino-live-mesh number, NOT the net48 Core engines. This blocks any beats-benchmark claim and any service exposure of a 3D density endpoint (truth criterion c).

Task: add C# harness rows (tools/Frahan.StonePack.Harness Validator.cs) that measure honest rho = sum(MeshPackItem.MeshVolume)/(A*H_used) at verified-zero interpenetration on ETH1100 for: the shipping GreedyMeshHeightmapPacker, and a Core AABB Dlbf3dMixedSizePacker fixture. The honest-density numerator already exists (MeshPackItem.MeshVolume, signed-tetra, shipped 2026-07-06).

Done when: RESULTS.md carries a measured C# 3D density row with the harness command that produced it, replacing the inherited Python figures. Size: M.

[libishm1]

Progress (v0.1.1-alpha): the honest-density measurement is now Rhino-free and shipped. MeshPackItem.MeshVolume (signed-tetra) is the honest numerator, and a headless C# bench runs the shipping GreedyMeshHeightmapPacker on the real ETH1100 subset: 16/16 placed, honest density 0.073 vs bbox 0.195 — a 2.68x bbox over-report (recorded in docs/results/RESULTS.md). Remaining before close: harness --pack3d rows per shipping packer at verified-zero interpenetration to fully replace the inherited Python figures. Keeping open for that.