Harden validate workflow checkout to resolve CodeQL cache-poisoning alert

[Copilot]

Summary

This PR addresses code scanning alert #2 by removing an untrusted workflow_dispatch ref input from the validate workflow. Checkout now uses a trusted GitHub context ref, eliminating the poisonable-step path without broad workflow changes.

• Scope
  • .github/workflows/validate.yml only.
• Change
  • Deleted workflow_dispatch.inputs.ref.
  • Switched checkout ref to github.ref.
• Before/after

# before
with:
  ref: ${{ inputs.ref || '' }}

# after
with:
  ref: ${{ github.ref }}

Type of change

• New capsule (models/<publisher>/<model>/<YYYY-MM-DD>/)
• New publisher (models/<publisher>/)
• New capability primer (docs/primers/<slug>.md)
• New glossary term (docs/GLOSSARY.md)
• CHANGELOG-only entry (announcement without capsule yet)
• Infrastructure / scripts / workflows
• Docs / README tidy-up

Capsule checklist (fill in for new capsules)

• Folder is models/<publisher>/<model>/<YYYY-MM-DD>/
• python scripts/validate.py passes
• references array in frontmatter has at least one model card
  and one docs link, all pointing at learn.microsoft.com when a
  canonical Learn page exists
• 1–3 concepts per notebook; extra concepts → extra notebooks
• Notebook sections numbered and alternate markdown ↔ code
• Every notebook ends with Your Turn to Explore, Summary,
  and References cells
• Notebook starts by verifying .env (points learner back to
  models/quickstart/ if unset)
• CHANGELOG.md has a row with matching Date + Model, exactly four
  cells wide, with the Date cell linked to the announcement
• Publisher README lists this model in its members table
• README.md Recently added block (Model / Release date /
  Capabilities) reflects the top 3
• No hype language ("revolutionary", "game-changing", "cutting-edge",
  "state-of-the-art" as a standalone claim, …)

Publisher / primer / glossary checklist

• Frontmatter validates
• Cross-links back to repo README and quickstart
• Grounded on learn.microsoft.com where a Learn page exists;
  provider docs only when necessary
• For a new capability primer: any existing capsule that already
  uses this tag now passes the crosslink check

Pedagogy self-review

• Voice is action-focused ("You'll deploy…", "Run this to see…")
• No marketing / brand adjectives (e.g. "powerful", "seamless",
  "world-class")
• Every learner-facing link resolves and points to a canonical
  source
• Uses "Microsoft Foundry" (never "Azure AI Foundry"); services
  like Azure OpenAI keep their real names

Reviewer notes

Code scanning alert #2 (actions/cache-poisoning/poisonable-step) was treated as a true positive and addressed by removing attacker-controlled ref selection in manual dispatch.