Added overloads to the (direct) array constructor functions in the stubs. These "bind" the type of the passed shape argument to the shape-type parameter of the numpy.ndarray. Note that this is only possible when the shape is passed as shape: int or shape: tuple[int, int], since e.g. the list type contains no information about its length (i.e. the number of dimensions).

This affects the type-signatures of:

• numpy.empty (numpy._core.multiarray.empty)
• numpy.zeros (numpy._core.multiarray.zeros)
• numpy.ones (numpy._core.numeric.ones)
• numpy.full (numpy._core.numeric.full)

implementation notes:

• For the sake of "DRY", I reduced some duplicated keyword-only parameters with typing.TypedDict and typing_extensions.Unpack, as defined in PEP 692.
• Because empty, zeros and ones have an identical signature (from a typing perspective), and each one has 12 overloads, I refactored them using a callable protocol, and decorated it with @typing.type_check_only.
• It's ok to use typing_extensions in the stubs, see TYP: Assume that typing_extensions is always available in the stubs #27132 .

This should be backwards-compatible, because the ndarray shape-type parameter is covariant, see #26081.