ugrid attribute is inconvenient about iodata

[tovrstra]

I'd suggest introducing a cube attribute in the IOData class, which will replace the current cube_data and ugrid. This cube would be an instance of the following type:

class Cube(NamedTuple):
    origin: np.ndarray
    axes: np.ndarray
    shape: np.ndarray
    data: np.ndarray  # replaces cube_data

This is the relevant part of the current docstring for the old attributes, which I'll adapt for the docstring of the Cube class:

    cube_data
        A (K, L, M) array of data on a uniform grid (defined by ugrid).

...

    ugrid
        A dictionary describing the uniform grid (typically from a cube file).
        It contains the following fields: ``origin``, a 3D vector with the
        origin of the axes frame. ``axes`` a 3x3 array where each row represents
        the spacing between two neighboring grid points along the first, second
        and third axis, respectively. ``shape`` A three-tuple with the number of
        points along each axis, respectively.

The old approach stems from certain data structures in HORTON2, which we do not have anymore in IOData, so there is no need to use exactly the same structure. (All relevant parts are still present.) The new approach is a little simpler and lighter and still works for Cube, CHGCAR and POTCAR files.

@PaulWAyers @FarnazH If you have any suggestions or comments, please let me know.