Add an interactive evaluation dashboard

## Summary

Add a read-only web dashboard for exploring Flowly evaluation runs and historical leaderboards. This is an optional follow-up to #38.

## Scope

- Browse persisted benchmark runs by suite and model
- Compare quality, latency, tokens, cost, tool success, patch applicability, and human acceptance
- Filter historical results and inspect scenario-level evidence
- Reuse the existing report store and metrics instead of creating a second evaluation data model

## Acceptance criteria

- [ ] The dashboard renders persisted reports and historical leaderboard data
- [ ] Model and run comparisons expose scenario-level metrics and regression status
- [ ] Empty, unknown-cost, and failed-run states are clear
- [ ] The UI is read-only and does not expose provider keys or other secrets
- [ ] Tests and user documentation cover the dashboard workflow

## Non-goals

- Running benchmarks from the dashboard
- Hosted/public report sharing (tracked separately)
- Automatic model routing recommendations (tracked separately)

Start by reading the existing report store and evaluation metrics, then review the related follow-up issue #38 to understand the existing workflow. Define the dashboard around persisted reports and historical leaderboards, keeping it read-only. Done means the acceptance criteria are met, including scenario-level comparisons, clear empty and failed states, tests, and user documentation.

Written by the indexing model from the issue text.