[imatiach-msft]

Summary

• add NumPy 1.x and 2.x compatibility coverage without changing the package version
• keep general runtime dependencies broad while requiring released ml-wrappers~=0.6.4, which contains the upstream NumPy compatibility work
• keep exact validated minimum-stack versions in test-only requirements rather than package metadata
• add latest and minimum dependency compatibility lanes for NumPy 1.26 and 2.x
• validate public TabularExplainer imports and NumPy NaN/Inf serialization
• restore the editable PR package after visualization dependencies install an older PyPI interpret-community
• stabilize legacy CI by isolating LightGBM estimators per transformation case, checking only platform-stable leading random-forest rankings, and pinning the last validated flake8-bugbear release

Dependency chain

• Add explicit NumPy 1.x and 2.x compatibility coverage microsoft/ml-wrappers#190 merged the NumPy compatibility changes
• ml-wrappers 0.6.4 was released to PyPI with those changes
• no additional prerequisite package release is required for this PR

Validation

• minimum stack passed with NumPy 2.0.2, Pandas 2.2.2, SciPy 1.13.0, scikit-learn 1.4.2, ml-wrappers 0.6.4, interpret-core 0.6.9, SHAP 0.46.0, and raiutils 0.4.2
• pip check passed for the minimum stack
• NumPy compatibility test passed against the editable PR source
• simulated the failing CI installation order: PyPI interpret-community 0.31.0 was replaced successfully by the editable 0.32.0 PR source, restoring reformat_importance_values
• package wheel built with broad runtime dependencies and ml-wrappers~=0.6.4
• focused legacy explainer tests passed with the CI dependency versions
• all PR checks pass: Python 3.9 full suite, four NumPy compatibility lanes, Python linting, CodeQL, and DCO
• workflow YAML and whitespace checks passed