fix issue 733

[meubleancien]

Fixes: #733

Removed thepandas<3.0.4 cap (PR #732), which was a segfault workaround; the real constraint was numba, since shap pulls in numba and every numbarelease declaresnumpy<2.5. Set pandas>=2.2.2,<4.0.0, numpy>=2.0.0,<2.5.0 (matching what numba actually supports), and raised the numbafloor to >=0.64.0 (first release supporting numpy 2.4.x); numba 0.66.0 also now guards the numpy version with a cleanImportError instead of segfaulting.
Verified against the reported segfault combo (pandas 3.0.3, numpy 2.4.6, numba 0.66.0, shap 0.52.0): 539 unit + 16 integration tests pass and end-to-end SmartExplainer.compile runs with no segfault.

Note: numpy stays effectively capped at 2.4.x by numba, so bump the numba floor and numpy ceiling together once a future numba adds 2.5 support.