Set column slices using an ndarray view of the underlying data

[taldcroft]

As discussed in #2994, in Python 2 setting a table Column slice does not go via __setitem__, but instead via __setslice__. Unless this method is overridden in the subclass then it ends up falling through to __array_finalize__, which is quite slow. This PR improves the speed by an order of magnitude or more for:

In [1]: from astropy.table import Column
In [2]: c = Column([1,2], name='a')
In [3]: timeit c[:] = 2
100000 loops, best of 3: 2.39 µs per loop

[embray]

Ack, I wish I'd remembered this sooner and saved a lot of trouble. I've had to use this workaround before but I'm slowly forgetting all the Python 2-isms :(

[taldcroft]

BTW, no new tests because slicing is well tested already in Table.

[mhvk]

@taldcroft - OK, thanks for adding that. The only change I would make is add the PR number to the relevant entry I had already made in CHANGES.rst for #2994. With that, this is ready to go.

[embray]

👍