feat(site): surface classified failure details

[SamJUK]

What changed

This updates the Astro frontend to surface the structured failure classifications we now record in result JSON.

It adds:

• version and baseline failure breakdown sections with grouped failure buckets
• compatibility vs harness style badges and summary counts
• affected variant chips for each bucket
• clickable affected variants that open the matching run logs
• a top-of-dialog failure callout in the logs modal for failed runs
• a homepage callout explaining classified failure signatures

Why

The existing UI made it hard to understand whether a failed run was a real compatibility problem or a harness issue without digging through raw logs.

This change makes the known failure classifications visible in the site so visitors and maintainers can identify likely root cause faster.

Impact

Visitors can now:

• see grouped failure reasons per version and per baseline page
• distinguish likely compatibility failures from harness failures at a glance
• jump straight from a failure bucket to the exact run logs for the affected variant

Validation

• pnpm --dir site build

[github-actions]

Result Comparison

Full diff

Result diff

Base: origin/master → Head: HEAD

No step status changes detected.