Statistics registry

Pure statistical computation (no Altair).

Backs the chart-annotation constructors in layers.py (notably add_comparisons). Holds the omnibus tests, hand-rolled post-hoc tests, effect-size functions, and the descriptive report builder. Nothing here imports Altair, so it is unit-testable in isolation.

The post-hoc tests scipy does not ship (Dunn, Nemenyi, Games-Howell) are implemented here from scipy primitives (rankdata, norm, studentized_range) rather than taking a dependency on scikit-posthocs (which would drag in statsmodels + seaborn + matplotlib).

clear_stats

def clear_stats() -> None: ...

Discard all pending statistical records queued by add_comparisons / add_correlation.

save() embeds only the records whose annotations appear in the chart being saved, so stale records never contaminate a save. But they do accumulate in memory across a long session — e.g. a notebook where you build many stats charts and display them without saving each. Call this to drop the pending queue.