refactor(lgrutil): coerce numpy types to builtins for np2 compat

[wpbonelli]

• test_lgrutil.py fails with numpy>=2.0.0rc1
• wrap numpy scalars with float() to avoid loss of precision, per https://numpy.org/devdocs/numpy_2_0_migration_guide.html#changes-to-numpy-data-type-promotion
• use builtin types in custom output data structures, not mixed builtins and np types — this seems preferable in general, and here in particular for lgr.get_exchange_data()

[langevin-usgs]

This seems a bit surprising to me that we have to explicitly wrap numpy scalars with int and float, but glad you were able to implement a fix.

[wpbonelli]

Numpy migration notes say this is expected for expressions where float32 is combined with the builtin float, this was the only one caught by a test but there are probably more cases where the change in promotion rules leads to slightly different results,. In general it may suffice to relax test comparisons but where we return non-numpy data structures (as here) it seems worth converting any np scalars to builtins for consistency