Make `DataSample` values generic to support shape typing

[redeboer]

Currently, values of the DataSample dict alias are NumPy arrays. A more backend-agnostic solution might be #131 and the new numpy.typing module, but an alternative might be to use a (possibly bound) TypeVar. This also might also improve deduced types in case there are scalar values in the data sample given to a Function/ParametrizedFunction (currently, this is marked as wrong by mypy/pyright).

Update below: #377 (comment)

[redeboer]

Since Python 3.11, there is TypeVarTuple, which you can use for variadic typing (PEP 646). NumPy does not (and will not?¹) be able to type shapes of arrays, but it is possible to do this externally using TypeVarTuple.

Footnotes

1. NumPy is very strict with its dependencies: it doesn't have any. This means it also does not have typing_extensions and therefore cannot type array shapes with variadics. ↩