feat: evals page — per-domain dashboard with trend lines and curation recommendations

[rajnavakoti]

Description

Once multi-domain and run comparison are in place, the Evals page should become a per-domain health dashboard that tracks coverage over time and recommends what to curate next. This is the north-star vision for the eval surface.

Acceptance Criteria

• Per-domain coverage dashboard with trend lines (coverage % over time)
• "What to curate next" recommendations driven by eval gap analysis (highest-impact MISSING/INCOMPLETE topics)
• Coverage threshold configuration per domain (e.g., "cash-payments needs 80% CLEAN to be operationally useful")
• Export eval report as shareable artifact (PDF or markdown)
• Curation effort estimation based on gap analysis (how many entities/hours to reach target coverage)

Out of Scope

• Automated curation (human reviews and approves all changes)
• Integration with external task management (Jira, Linear)