Fix test_slicing_errors failing for older NumPy versions

[nils-werner]

Prior to NumPy 1.13.0, you would see a VisibleDeprecationWarning but no exception when using illegal indexing values. This makes test test_slicing_errors[index6] fail.

__________________________________ test_slicing_errors[index6] ____________________________________

index = [True, False, False]

    @pytest.mark.parametrize('index', [
        (Ellipsis, Ellipsis),
        (1, 1, 1, 1),
        (slice(None),) * 4,
        5,
        -5,
        'foo',
        ([True, False, False]),
    ])
    def test_slicing_errors(index):
        s = sparse.random((2, 3, 4), density=0.5)
        x = s.todense()

        try:
            x[index]
        except Exception as e:
            e1 = e
        else:
>           raise Exception("exception not raised")
E           Exception: exception not raised

sparse/tests/test_core.py:473: Exception
===================================== warnings summary ==============================================
sparse/tests/test_core.py::test_slicing_errors[index6]
  /Users/nwerner/Arbeit/mdct/sparse/sparse/tests/test_core.py:469: VisibleDeprecationWarning: boolean
  index did not match indexed array along dimension 0; dimension is 2 but corresponding boolean dimension is 3
    x[index]

-- Docs: http://doc.pytest.org/en/latest/warnings.html

Either the test should be skipped for NumPy < 1.13.0, or we should bump the dependency version.

[mrocklin]

I'm in favor of skipping or xfailing the test rather than require this numpy version.

[nils-werner]

Makes sense, I replaced the requirement with a pytest.mark.skipif decorator.

[hameerabbasi]

I agree with @mrocklin here. Since we're emulating the latest Numpy, but not really using anything from it, it makes sense to skip the test rather than bump the version.

[nils-werner]

This should be ready to merge.