Add a reusable benchmark card template - #61

Description of Changes

Adds a benchmark-level documentation template covering identity and lifecycle, intended decisions, task composition, taxonomy coverage, protocol, measures and uncertainty, validation evidence, safety boundaries, and governance. It explicitly separates benchmark-level evidence from Unitxt-style executable task cards and includes an abbreviated MedFailBench example showing how to state a high-stakes clinical boundary.

The evaluations landing page now links to the template and exposes it as a child page.

Related Issues

Closes #24

Testing Performed

• Verified site builds successfully (make view-local requires Ruby >=3 because the current Nokogiri dependency no longer resolves on the macOS Ruby 2.6 runner used here)
• Checked external-link target declarations with ./check-external-links.sh (0 findings)
• Checked the new page with markdownlint-cli2 (0 errors; repository-compatible MD013/MD025/MD060 exemptions)
• Parsed all Markdown front matter with Ruby YAML and ran git diff --check

Code Changes

• docs/evaluations/benchmark-card-template.markdown
• docs/evaluations/evaluations.markdown

Preview

Review the field coverage, especially the distinction between executable task cards and benchmark-level composition, validity, and decision-boundary evidence.

Checklist

• I have read the contributing guide and used a DCO sign-off
• I have updated related documentation
• I have included a worked boundary-language example

Additional Context

This is intentionally a documentation POC. It links to Unitxt task cards and the EvalEval catalog rather than duplicating their executable or standardized records.