docs: fix kl_divergence and chi_square audit arguments and semantics

[shinichitazawa]

Description

The examples for two statistical audits in docs/concepts/audits.md use argument names that the built-in audits do not accept. Copying them fails at render time with Macro variable 'source_column' is undefined.

Audit	Documented	Implemented (sqlmesh/core/audit/builtin.py)
kl_divergence	column, target_column	source_column, target_column
chi_square	column, target_column	column_a, column_b (plus dependent)

The chi_square description also said the audit "ensures that the chi-square statistic ... does not exceed a critical value". The implementation does the opposite by default:

HAVING NOT @IF(@dependent, chi_square > @critical_value, chi_square <= @critical_value)

With dependent := true (the default) the audit asserts that the two columns are dependent and fails when the statistic does not exceed the critical value. This PR documents the dependent argument and both behaviours, updates the example text accordingly, and fixes the "statistic0" typo. It also aligns the kl_divergence usage comment in builtin.py with its arguments.

Note: even with the corrected argument names, kl_divergence currently fails at run time on DuckDB 1.5.6 and Trino 483 (LEFT JOIN ... USING (@source_column) requires both columns to share a name, and when they do, num_rows becomes ambiguous). That is an implementation issue outside the scope of this docs change.

Test Plan

Ran each audit with SQLMesh 0.236.3 on DuckDB 1.5.6 against a single hourly interval (INCREMENTAL_BY_TIME_RANGE model, ~329k rows):

• chi_square(column := ..., target_column := ...) (current docs): fails with Macro variable ... is undefined
• chi_square(column_a := src_ns, column_b := dst_ns, critical_value := 6.635): passes (statistic ~889,337, columns are dependent)
• Same with dependent := false: fails with 1 row failed, confirming the documented semantics
• kl_divergence(column := ...) (current docs): fails with Macro variable 'source_column' is undefined; with source_column := the macro resolves (run-time failure described above)

ruff check / ruff format --check pass on builtin.py (comment-only change). make fast-test was not run since no code paths changed.

Checklist

• I have run make style and fixed any issues (ran ruff check and ruff format --check on the changed Python file; docs are not covered by pre-commit hooks)
• I have added tests for my changes (if applicable) — not applicable, docs and a comment only
• All existing tests pass (make fast-test) — not run, no code paths changed
• My commits are signed off (git commit -s) per the DCO

🤖 Generated with Claude Code