Package 10.8.3: 32/48/64 full-grid convergence and periodic surface topology

[aliorhun]

Package 10.8.3 — 32³/48³/64³ real-map convergence + topology gate

The gate checks two distinct numeric renderings from the same production Worker:

• Current default: ~40 intervals per axis (64³ source → effective stride 2).
• Validation-only full-grid research comparison: up to 64 intervals per axis, stride 1 at 32³, 48³ and 64³.
• Both modes keep the 75,000-triangle cap, with truncation/empty meshes treated as CI failures. The final production rendering and science algorithm are unchanged.
• Map sources: pinned Gemmi 5i55 real partial monoclinic CCP4 and 5wkd real C2 MTZ with independently regenerated FWT/PHWT and DELFWT/PHDELWT maps.
• Physical Ų areas, thresholded voxel quadrature ų volume, mesh triangle counts, stride, and interior-vs-boundary unmatched edges are reported. A signed synthetic sphere crossing a skew periodic seam verifies correct periodic boundary matching (4,368 triangles / 0 unmatched seam edges).
• Genuine non-manifold edges, interior open edges, physical degeneracies and cap overruns fail CI. Open domain edges for partial experimental maps are reported and accepted. Orientability discrepancies are reported, not silently repaired.

Actual measured 48³ → 64³ full-grid sensitivity

Real map	Surface area change	Thresholded occupied volume change
5i55 partial CCP4	5.84%	0.94%
5wkd FWT/PHWT	62.33%	40.85%
5wkd DELFWT/PHDELWT	52.71%	0.40%

Scientific interpretation: real 5wkd surfaces are NOT proven converged at 64³, even though the scientific validation gate passed. The gate validates the pipeline, geometry and reproducible measurements; it does not impose an unjustified monotonicity condition on experimentally derived maps. A reproducible non-node threshold is chosen near the signed-density peak percentile across the compared grids to prevent isovalues exactly equal to a sampled vertex. No falsely labelled independent CHGCAR/CUBE/XSF corpus, Hausdorff guarantee, or multi-GPU benchmark.

Next: inspect non-convergent FWT/difference surfaces, compare an independent mesher and consider quality-budget tradeoffs on test datasets before changing the UI default.

Please leave unmerged until both native/Gemmi/scientific and Chromium/WebGL CI gates pass.