Particle Statistics: AdStat integration, ingress/export, opt-in ordering, docs

[jacobson30-bot]

Brings the Particle Statistics tool (AdStat-powered) to a committable, well-tested state.

Highlights

• Backend / correctness (adstat_adapter.py): density-aware nearest-neighbour scale; warn on heterogeneous pooled regions; hardened sandbox-preview reach into an AdStat internal.
• Data ingress → one shared, poolable feature-set store: app-level FeatureSetStore shared by image viewers + Feature Counting; Feature Counting Send to Particle Statistics; Load points from disk… importer (CSV position tables + ProbeFlow JSON) with a prefilled calibration dialog; Save feature sets…. Un-strands feature_counting_to_particle_table and FeatureSetStore.load/save.
• Result export: per-statistic curve CSVs + verdicts CSV + full JSON via a top Export menu.
• Opt-in local-order statistics (ψ4/ψ6, angular g(r,θ)) — off by default — with a known-answer validation suite proving they reject true lattices while random data stays consistent.
• Tutorial drawer redesign: 3-panel Previous | current lesson | Next navigation that never dead-ends; fixed exit/landing leaving the dialog behind Browse.
• Definitions: a Particle Statistics tab in the shared ProbeFlow Definitions document — a rigorous, defensible reference (null models, statistics, the ERL envelope test, honest caveats); the dialog's Definitions menu opens it. Removed the isolated in-dialog tab.
• Tidy: removed the redundant standalone AdStat sandbox dialog; pointed the Model simulations command at the sandbox mode; documented the mode vocabulary; docs + an honest maturity note throughout.

Testing

Full repo suite green with a Qt-enabled offscreen environment: 2604 passed, 3 skipped, ruff clean. New GUI-free suites: test_point_table_io, test_adstat_export, test_adstat_validation.

Notes

• Single comprehensive commit; it bundles the pre-existing (previously uncommitted) AdStat adapter work with this round, since they interleave across shared files.
• Particle Statistics is the newest, least user-tested area — the docs and in-app reference say so. A real visual pass over the new tutorial drawer and Definitions tab is still worth doing.

🤖 Generated with Claude Code