feat(examples): add compliance-lint GitHub Action reference (Phase 3 of #1172)

[christso]

Summary

Phase 3 of #1172 — additive, no core changes.

Ships a reference GitHub Action that uses the agentv-compliance skill (landed in Phase 1, #1178) to lint governance: blocks in changed *.eval.yaml files on pull_request events.

Files:

• examples/governance/compliance-lint/compliance-lint.yml — reusable workflow; triggers on **/*.eval.yaml path changes; posts PR comment; exits non-zero on violations
• examples/governance/compliance-lint/script/lint.py — extracts governance blocks, calls claude-haiku-4-5 with the skill loaded, returns { pass, violations[] } JSON
• examples/governance/compliance-lint/README.md — adoption guide (5-minute setup, cost estimate, mandatory-gate instructions, customization)

Design choices:

• Python + anthropic SDK: lightest CI setup, no Bun/Node dependency on CI runners
• claude-haiku-4-5: ~$0.03 per 10-file PR (well under the 5¢ target documented in README)
• Skill loaded once per run, passed as system prompt to all block lints in that run
• PR comment is best-effort (skipped if GITHUB_TOKEN/PR_NUMBER/REPO are absent)
• Not wired into agentv's own CI (opt-in only, as specified in issue non-goals)

Test plan

• Pre-push hook passed: Build ✓ Typecheck ✓ Lint ✓ Test ✓ Validate eval YAML ✓
• Manually verified lint.py against llm01-prompt-injection.eval.yaml (passes with no violations)
• Verified malformed block produces clear violation report (risk_tier super_high → risk_tier_value violation with suggestion)
• README documents adoption in under 5 minutes, cost estimate included
• No mandatory CI gate added to agentv repo itself (non-goal per issue)

Closes #1172 (Phase 3, final)