Fix InsightEngine URL-less result deduplication

[3em0]

Summary

• replaces the URL-less result dedupe key that previously used title_or_content[:100]
• preserves URL-based deduplication for results that have a URL
• adds regression coverage for URL-less results that share the same first 100 characters but differ in full content

Fixes #688

Tests

• python -m pytest tests/test_insight_deduplication.py