Per-column field rules — caps, drops, scalar preservation in compact view

[DEVunderdog]

Parent

#268

What to build

Apply the compact field rules inside each flagged column's detail section. Drop: histogram bins, total_rows from ColumnMissingnessProfile, memory_breakdown (already excluded at dataset level — confirm nothing leaks here), missingness_matrix (already excluded at dataset level — same). Cap: top_values to 3 entries for both NumericStats and CategoricalStats. Keep in full: all scalar fields including redundant pairs (std + variance, standard_null_ratio + effective_null_ratio, mean_median_ratio), all enum tags and flags, all PercentileSnapshot values (7 scalars), all BimodalStats fields, all RowMissingnessDistribution fields, correlated_with (column names list, no cap).

Acceptance criteria

• Histogram bins do not appear in any flagged column's detail section
• total_rows does not appear in any column's missingness subsection
• top_values sections contain at most 3 entries for numeric and categorical columns
• All scalar fields are present: mean, median, std, variance, skewness, kurtosis, min, max, mode, mode_frequency, mean_median_ratio, effective_null_ratio, standard_null_ratio, effective_null_count, standard_null_count, all 7 percentile values, outlier_density, tail_asymmetry_ratio
• correlated_with appears in full (no cap on column name list)
• BimodalStats fields appear in full when NumericFlag.Bimodal is set
• Unit tests assert absence of dropped fields and presence of all scalars using a NumericStats-rich fixture

Blocked by

• Blocked by Two-tier column rendering — clean threshold, flagged detail sections, severity ordering #271