Fix null counts exceeding per-file record counts

[keemgdeok]

Summary

Treat a top-level column as having inconsistent metrics when a data file reports more nulls than records, independently of whether its value count is present. The affected null-ratio rule now returns UNKNOWN / INCONSISTENT_METRICS; row-count evidence and unrelated columns remain unaffected.

Regression coverage

• Missing or inflated value counts with null counts exceeding file rows.
• Positive null counts in a zero-row file.
• Valid all-null, zero-null, and zero-row boundaries.
• Per-file contradictions hidden by the total row count across other files.
• Row-count and unrelated-column results remain valid.

Validation

• Confirmed four regression inputs failed before the production fix.
• Java 17 Docker build and formatting checks passed; all 45 tests passed.
• Repeated the original 10-row / 20-null reproduction against the rebuilt JAR: null ratio returns UNKNOWN / INCONSISTENT_METRICS, while row count returns PASS.
• Packaged doctor --demo passed with the no-content-file guard.
• git diff --check passed.

Production change is limited to a three-line consistency check. No API, dependency, release-version, or E2E-environment changes.