Mypy and NumPy 2.x Require Stricter Types

[vitenti]

When using mypy with NumPy 2.x, stricter type definitions are necessary. For example, functions that return an array from np.linspace often produce an NDArray with dtype np.floating[Any]. This becomes an issue when the function signature specifies a return type like -> npt.NDArray[np.float64], as it causes a type error.

Should we relax the type hint to allow any floating-point type (e.g., converting everything to np.floating[Any] or using a defined alias), or should we enforce strict adherence to np.float64?

[marcpaterno]

If we are looking for the best reproducibility, and also if we are interfacing with compiled C or C++ that uses double on any modern architecture, then I think we should stick with np.float64. This is the clearest expression of what I think we mean: a 64 bit IEEE-754 floating point type, that can be exchanged between different languages without requiring conversions.