feat(governance): add regional bias audit framework for model fairness

[Goldokpa]

Summary

• Adds BiasAuditor class for evaluating model fairness across geographic regions
• Implements three fairness metrics: demographic parity, equalized odds, predictive parity
• Creates scripts/audit_model.py CLI tool for running audits
• Adds notebooks/07_bias_audit.ipynb with visualization examples
• Supports regions: Amazon, Congo, Southeast Asia, Boreal

Key Features

• Per-region metrics: IoU, F1, precision, recall, TPR, FPR
• Fairness scoring: Detects disparate performance across regions
• CI/CD integration: check_fairness_gate() returns True/False for pipeline gating
• Automated recommendations: Suggests improvements for underperforming regions

Usage

# Run full audit
python scripts/audit_model.py --model models/best_model.pth --regions amazon,congo

# CI gate mode (exit 1 if fails)
python scripts/audit_model.py --model models/best_model.pth --ci-gate --threshold 0.85

Test plan

• Run python -c "from climatevision.governance import run_bias_audit; print('OK')"
• Execute python scripts/audit_model.py --list-regions
• Execute notebooks/07_bias_audit.ipynb cells
• Verify reports save to outputs/bias_reports/

Closes #23

🤖 Generated with Claude Code

[obielin]

Thanks for picking this up while I was finishing the explainability work — the regional bias auditor with demographic-parity / equalized-odds support is exactly what the governance pipeline needs upstream of the CI gate I shipped in #39.

After the rebase the PR landed cleanly on top of my anomaly detector / audit logger / model card / LLM reporter — governance/__init__.py re-exports all of them coherently. The fairness gate function plugs straight into scripts/governance_ci_gate.py. Approving and happy to take ownership going forward.