`test_nan_propagation` works incorrectly for `cumulative_sum`

[ndgrigorian]

Per spec,

if include_initial is False, the returned array must have the same shape as x.

however, test_nan_propagation expects the result to have an empty shape (i.e., be a scalar).

func_name = 'cumulative_sum', x = usm_ndarray([nan], dtype=float32), data = data(...)

    @reproduce_failure('6.102.4', b'AXicY2BkQAcAACUAAg==')
    @pytest.mark.parametrize(
        "func_name", [f.__name__ for f in category_to_funcs["statistical"]]
    )
    @given(
        x=hh.arrays(dtype=xps.floating_dtypes(), shape=hh.shapes(min_side=1)),
        data=st.data(),
    )
    def test_nan_propagation(func_name, x, data):
        func = getattr(xp, func_name)
        set_idx = data.draw(
            xps.indices(x.shape, max_dims=0, allow_ellipsis=False), label="set idx"
        )
        x[set_idx] = float("nan")
        note(f"{x=}")

        out = func(x)

>       ph.assert_shape(func_name, out_shape=out.shape, expected=())  # sanity check
E       AssertionError: out.shape=(1,), but should be () [cumulative_sum()]

[asmeurer]

Fixed at #262

[ndgrigorian]

Since #262 was merged I am closing this as completed.

Thank you!