Add membership inference privacy gates to ai-data-privacy

[alejandrorivas-pixel]

Summary

• Adds a dedicated membership-inference and privacy-attack evidence gate to ai-data-privacy
• Covers confidence/logit/top-k exposure, embedding/RAG similarity-score and document-ID leaks, label-only residual risk, query-budget controls, and differential privacy evidence quality
• Updates output categories, privacy control summary, severity guidance, pitfalls, version, and references
• Adds focused benign/vulnerable fixtures for synthetic label-only N/A, confidence-score membership leakage, and RAG similarity-score membership leakage

Related issue

Closes #1213

Test Cases Added

• skills/ai-security/ai-data-privacy/tests/benign/synthetic-label-only-membership-not-applicable.md
• skills/ai-security/ai-data-privacy/tests/vulnerable/confidence-score-membership-leak.py
• skills/ai-security/ai-data-privacy/tests/vulnerable/rag-similarity-score-membership-leak.py

Validation

• git diff --check
• python3 -m py_compile on Python fixtures
• Frontmatter required-field sweep across skills/ and roles/
• Prompt-injection scan equivalent to the repository workflow
• Markdown fence balance check: 16 fences, balanced
• Marker checks for membership-inference section, predict_proba, similarity_search_with_score, Not Applicable, Not Evaluable, AML.T0024.000, label-only reference, and fixture-specific expected signals
• Reference URL checks returned HTTP 200 for OWASP, NIST AI RMF, arXiv membership-inference sources, and MITRE ATLAS base site

Bounty

Bounty target: Improver Moderate ($100) if accepted. Payment details can be provided privately after maintainer acceptance.