Fix comparison with NumPy of slicing with negative step

[venkat0791]

The existing documentation states that the behavior of slicing is the same as in NumPy except when step < -1, implying that the behavior is the same when step = -1. But this is not true:

In [1]: import numpy as np

In [2]: x = np.arange(10)

In [3]: x[2 : 5 : -1] # Analogous slice in ndarray: array![4, 3, 2]
Out[3]: array([], dtype=int32)

In [4]: x[5 : 2 : -1] # Analogous slice in ndarray: array![]
Out[4]: array([5, 4, 3])

So step < -1 should be replaced by step < 0 in the documentation.

There are some further differences in slicing behavior with negative step, having to do with the default values for start and end:

In [5]: x[: 7 : -1] # Analogous slice in ndarray: array![6, 5, 4, 3, 2, 1, 0]
Out[5]: array([9, 8])

In [6]: x[7 : : -1] # Analogous slice in ndarray: array![9, 8, 7]
Out[6]: array([7, 6, 5, 4, 3, 2, 1, 0])

[nilgoyette]

Thank you @venkat0791 for fixing this wrong information. The CI tests are failing, but it's not related to this MR so I'll merge anyway.

I don't remember using negative slicing in ndarray so I learned something :)