Implement automated model security testing and adversarial robustness

[gelluisaac]

Description

    Build security testing framework for ML models including adversarial attack simulation, model extraction detection, and poisoning detection.

    ## Files to modify
    - astroml/security/adversarial/attacks.py
    - astroml/security/adversarial/defenses.py
    - astroml/security/model_extraction.py
    - astroml/security/poisoning_detection.py
    - astroml/api/routers/model_security.py

    ## Out of scope
    - Refactoring unrelated modules
    - Changing public API contracts
    - Major version bumps or breaking changes
    - Performance optimizations not directly related to the issue

    ## Acceptance criteria
    - [ ] Code changes implemented as per procedure steps
    - [ ] All new code has complete type hints
    - [ ] Unit tests added/updated with >90% coverage for changed code
    - [ ] All tests pass: pytest tests/
    - [ ] Linting passes: pre-commit run --all-files
    - [ ] Documentation updated if needed
    - [ ] Changes reviewed and approved

    ## Procedure
    1. Implement adversarial attack generation (FGSM, PGD, CW)
    2. Build adversarial defense mechanisms
    3. Add model extraction detection
    4. Implement data poisoning detection
    5. Create model security scoring
    6. Build security testing pipeline
    7. Add security report generation
    8. Test with various attack scenarios