data_diff hardening (redaction, limits, validation, docs, refactor) #1680

Follow-ups from the v0.5.22 and v0.6 release reviews of data_diff, collected in one place.

Items

• data_diff: add dedicated unit test file for pure functions #718: Dedicated unit tests for the pure functions.
• data_diff: PII redaction / opt-in for sample diff rows #719: PII: sample differing rows (raw values) reach the model and transcripts; add enforced suppression or redaction. feat: data_diff sample-row redaction / include_values opt-in #729 proposed ALTIMATE_DATA_DIFF_INCLUDE_VALUES plus an include_sample_values parameter.
• data_diff: configurable MAX_STEPS and wall-clock timeout #720: Configurable MAX_STEPS and a wall-clock timeout, with telemetry on cap hits.
• data_diff: fuzzy column-name suggestions on errors #721: Fuzzy column-name suggestions on errors, using discovered columns.
• data_diff: where_clause grammar validation #722: where_clause grammar validation, including UNION rejection (write and multi-statement protection already shipped).
• docs: data-parity guide page with screenshots #723: Data-parity guide: screenshots and an explicit decision tree.
• feat: data_diff audit log #730: Structured audit log of comparisons, with rotation.
• feat: data_diff row-count ceiling (max_rows guard) #731: Optional row-count ceiling (max_rows) with an actionable rejection.
• refactor: split data-diff.ts (1118 lines) into focused modules #732: Split data-diff.ts into focused modules without behavior change.

The linked issues hold the original reports and reproductions. Most are closed in favor of this tracker; any child with an open fix PR stays open until that PR lands.