Add a pinned LintLang gate for authored SKILL.md files

Desired Behavior

Add a deterministic, zero-LLM check for authored SKILL.md files, using a pinned release of LintLang. LintLang flags statically-detectable instruction-language problems (unbounded retry loops, missing priority ordering, ambiguous conditionals) with no network or model calls.

Proposed shape, as a new lintlang-skill-gate job in .github/workflows/tests.yml:

• Select every SKILL.md under skills/ and plugins/ (79 files today).
• Run uvx --from lintlang==0.6.0 lintlang scan <files> --fail-on fail.
• Print all findings, but fail only on HIGH/CRITICAL. No continue-on-error, no baseline file.

Current default branch (1bad294): 79 scanned, 0 FAIL, 0 HIGH/CRITICAL, exit 0. So the gate starts green and only blocks regressions.

This is additive to sync_extensions.py --check and validate-claude-code; it checks instruction-language structure, not runtime behavior or safety.

Acceptance Criteria

• tests.yml gains a lintlang-skill-gate job pinned to an exact LintLang release.
• REVIEW-tier findings are logged but do not fail the job; HIGH/CRITICAL do.
• No skill content, manifests, or lockfiles change.
• Existing Tests and Check Extensions jobs stay unmodified and green.

Start in .github/workflows/tests.yml by reviewing the existing Tests and Check Extensions jobs, then add the proposed lintlang-skill-gate job for SKILL.md files under skills/ and plugins/. Run the pinned LintLang command against the current 79 files and confirm REVIEW findings are logged while HIGH/CRITICAL findings fail the job, with existing jobs unchanged.