feat(skill): add agentv-compliance skill for governance block authoring and lint (Phase 1 of #1172)

[christso]

Summary

Phase 1 of #1172 — additive only, no core changes.

Creates plugins/agentv-dev/skills/agentv-compliance/ with 7 reference files that teach AI agents how to author and lint OWASP / MITRE ATLAS / EU AI Act / ISO 42001 governance blocks in *.eval.yaml files.

Structure:

• SKILL.md — top-level skill description; dual-mode (interactive authoring + non-interactive CI lint)
• references/owasp-llm-top-10-2025.md — LLM01–LLM10 with descriptions and source link
• references/owasp-agentic-top-10-2025.md — T01–T10 agentic-AI categories
• references/mitre-atlas.md — Common AML.Txxxx technique IDs with OWASP cross-references
• references/eu-ai-act-risk-tiers.md — Four risk tiers with article references
• references/iso-42001-controls.md — Curated ISO 42001:2023 controls for AI eval
• references/governance-yaml-shape.md — YAML shape, merge semantics, two complete real-suite examples
• references/lint-rules.md — 7 rules (known_key, ID format per framework, risk_tier values, array_not_empty) that Phase 3's Action uses to produce structured violation reports

Test plan

• All 8 skill files authored from scratch using existing suite examples as ground truth
• governance-yaml-shape.md examples copied verbatim from llm01-prompt-injection.eval.yaml and destructive-git.eval.yaml
• lint-rules.md covers all fields currently enforced by the validator being removed in Phase 2
• Pre-push hook passed: Build ✓ Typecheck ✓ Lint ✓ Test ✓ Validate eval YAML ✓
• No core changes — zero risk of regression

Closes #1172 (Phase 1)