feat: add compliance framework tagging for AI red teaming

[rdheekonda]

Add OWASP, ATLAS, SAIF, and NIST compliance tags to attacks and transforms. Attacks tagged with core jailbreak technique (LLM01), transforms tagged with specific vulnerability categories. Includes comprehensive test coverage.

Key Changes:

• Attacks show core technique (LLM01), transforms add vulnerability categories
• No tag duplication when attacks use transforms
• Clean separation: attack vector vs. vulnerability target

Added:

• dreadnode/airt/compliance/ - OWASP, ATLAS, SAIF, NIST framework definitions
• tag_attack() and tag_transform() helper functions
• 45 unit tests for compliance tagging (3 new test files)
• Compliance tags to 4 jailbreak attacks (TAP, GOAT, Crescendo, prompt)
• Compliance tags to 11 transform modules (130+ transforms)

Changed:

• Attack base class: added compliance_tags field
• Transform base class: added compliance_tags parameter
• Removed global variables, replaced with @functools.lru_cache
• Export compliance framework from dreadnode.airt

---

Generated Summary:

• Introduced compliance framework tagging for AI red teaming in the new compliance module.
• Added various compliance tags for OWASP, NIST, ATLAS, and SAIF frameworks:
  • Implemented tagging mechanism via tag_attack and tag_transform functions.
  • Integrated compliance tagging into existing attack methods (e.g., goat_attack, crescendo_attack, prompt_attack, tap_attack).
• Updated Transform class to accept compliance tags and modified transform methods to include them.
• Enhanced attack classes with a new compliance_tags dictionary for better traceability.
• Potential impact:
  • Improves visibility and traceability of attacks concerning industry standards.
  • Facilitates compliance audits and risk assessments.
  • Users can leverage compliance tags directly in attack metadata for better analysis.

This summary was generated with ❤️ by rigging