feat(engine): risk scoring by privacy exposure level

[martsokha]

DEVELOPER.md Section 5.1: documents and datasets should be scored by aggregate privacy exposure level, enabling organizations to prioritize review effort toward the highest-risk content.

• Define risk scoring model (entity count, type severity, confidence)
• Compute score during pipeline execution
• Include score in pipeline output
• Add integration test

Reference: docs/DEVELOPER.md Section 5.1