Be more precise when type hinting numpy floats

Numpy 2 deprecates the `np.float_` alias for `np.float64`, and since we often use `np.float32` we should type hint better to indicate when the variables/return types might be `np.float32` vs `np.float64` .

Start by locating type hints that use np.float_, np.float32, or np.float64, then inspect the surrounding annotations and their callers. Done means the affected variables and return types distinguish float32 from float64 and no longer rely on the deprecated np.float_ alias.

Written by the indexing model from the issue text.