feat: enriched recommendations with attack scenarios and likelihood (#90)

[reloadfast]

Summary

Implements issue #90 — Explain Like I'm Human (ELIH) mode for recommendations.

Changes

Backend:

• Added attack_scenario and likelihood fields to _Advice dataclass (default "" for backwards compat)
• Populated all 27 catalogue entries with plain-language exploitation narratives and realistic home network likelihood assessments
• Added attack_scenario and likelihood nullable columns to Recommendation ORM model
• Added schema migrations in _migrate_schema() for existing databases
• generate_recommendations() now writes both fields from the catalogue
• RecommendationOut and _rec_to_out() expose both fields via the API

Frontend:

• Added attack_scenario: string | null and likelihood: string | null to TypeScript Recommendation interface
• RecommendationDetailPage now renders two expandable sections below the remediation steps:
  • 🎯 How could this be exploited? — plain-language attack narrative
  • 📊 How likely is this on a home network? — realistic risk assessment
  • Both sections are conditionally rendered (only shown when fields are non-empty)

Testing

• All 241 existing tests pass
• Ruff and Prettier checks pass

Closes #90