feat(governance): add SHAP explainability for segmentation predictions

[Goldokpa]

Summary

• Adds src/climatevision/governance/ module with SHAP-based explainability
• Implements SHAPExplainer class using DeepExplainer with gradient fallback
• Adds /api/explain and /api/explain/{run_id} API endpoints
• Creates notebooks/06_explainability.ipynb with visualization examples
• Adds shap>=0.42.0 to requirements.txt

Key Features

• Band-level attribution: Shows which spectral bands (Red, Green, Blue, NIR) drove the prediction
• Spatial importance: Heatmap showing which image regions mattered most
• API integration: REST endpoints for generating explanations on-demand

Test plan

• Run python -c "from climatevision.governance import SHAPExplainer; print('OK')"
• Start API and test POST /api/explain endpoint
• Execute notebooks/06_explainability.ipynb cells
• Verify heatmap generation saves to outputs/explanations/

Closes #22

🤖 Generated with Claude Code