Build the agent-eval UI homepage route #429

Part of #428.

This issue tracks building / as the entry point for browsing benchmarks, experiments, scenarios, and runs in agent-eval ui and the exported static site.

The experimental homepage already lists benchmarks, experiments, and scenarios. We should build on that rather than start a separate implementation in website.

Scope

• Show an overview of benchmarks, experiments, scenarios, and recent runs, with links to their collection and detail pages.
• Keep resource names readable while using resource IDs in links.
• Handle empty collections and projects without saved runs, and distinguish missing data from failures to load data.
• Use a bounded overview rather than rendering every historical run on the homepage.
• Use Primer components and design tokens, support light/dark mode, and provide accessible headings and navigation.
• Verify navigation in local preview and static export, including a non-root deployment base path.

High-level statistics are a possible follow-up, not a requirement for this issue.

Related work

The collection/detail route issues can proceed in parallel. Integrate their links as they land; the runs issue owns shared result discovery. The homepage does not need to wait for all detail pages before implementation starts.

Starting point

packages/agent-eval/src/ui/app/page.tsx:1-55 currently lists the three configuration collections and links to plural detail URLs.