ci: add docs metadata health check

[omarespejel]

Summary

Add a lightweight live metadata health check for representative docs.starknet.io pages.

This complements the existing llms.txt validation and discovery endpoint checks. The goal is to catch deployed-site regressions that affect search, AI agents, and social/link previews without burdening normal docs content edits.

What This Checks

For key docs pages across homepage, build, protocol, reference, and operations sections, the script checks:

• HTTP 200 and text/html response
• non-empty page title
• canonical URL matches the public docs URL
• no noindex directive in meta robots or X-Robots-Tag
• presence of Open Graph/Twitter title metadata
• presence of JSON-LD structured data

Meta descriptions are currently reported as non-blocking warnings because several existing docs pages do not emit them yet. This keeps the check useful without turning existing metadata debt into a blocking gate.

Workflow Behavior

• Runs manually via workflow_dispatch
• Runs weekly on Monday
• Runs on PRs only when this health-check script or workflow changes
• Does not run on normal docs content edits

Examples Of Regressions This Catches

• A key docs page is accidentally marked noindex.
• A platform/CDN change removes canonical links from rendered docs pages.
• A representative docs page stops returning HTML or starts returning an error.
• Structured data or preview title metadata disappears from rendered pages.

Validation

Ran locally:

python3 -m py_compile scripts/check_metadata_health.py
python3 scripts/check_metadata_health.py
python3 scripts/check_metadata_health.py --base-url file:///tmp; test $? -eq 1
python3 scripts/check_metadata_health.py --base-url https://docs.starknet.io/foo; test $? -eq 1
python3 scripts/validate_llms_txt.py

Staged first in omarespejel/starknet-docs and addressed AI-review feedback from CodeRabbit and Qodo before opening this upstream PR.