Correlation section — top-5 Pearson + Spearman per column, Target Analysis, Sentinels

[DEVunderdog]

Parent

#268

What to build

Replace the full N×N correlation matrices in to_markdown() with compact top-5 views, and add the Target Analysis and Sentinels sections. For feature correlations: within each flagged column's detail section, show the top-5 highest absolute Pearson correlations and top-5 highest absolute Spearman correlations (column name + value, sorted descending by absolute value). Add a ## Target Analysis section showing, for each declared target column, the top-5 absolute Pearson and top-5 absolute Spearman correlations per feature column. Add a ## Sentinels section rendering numeric_sentinels and string_sentinels unchanged.

Acceptance criteria

• Feature correlation entries in a column detail section contain at most 5 Pearson and 5 Spearman entries, sorted by descending absolute value
• The full N×N Pearson and Spearman matrices do not appear anywhere in to_markdown() output
• ## Target Analysis section appears when result.dataset.target_correlations is non-empty, showing top-5 Pearson + top-5 Spearman per feature per target
• ## Sentinels section renders numeric_sentinels and string_sentinels from the result
• Unit tests cover: result with 10-column correlation matrix (assert ≤ 5 entries appear per column); result with target correlations (assert top-5 ranking is correct); result with sentinels (assert both dicts appear)
• to_markdown() is now complete per ADR-0040 — the full compact contract is satisfied

Blocked by

• Blocked by Per-column field rules — caps, drops, scalar preservation in compact view #272