Rewrote topolgical_sort as a generator:
* It no longer accepts reverse or nbunch arguments
* It is slightly faster.
Added lexicographical_topological_sort, which accepts a key.
Some tests for topological sort were removed because they relied on
CPython's particular dict element ordering.
Added utils from more_itertools:
* These were necessary for working with the new generators.
* "pairwise" is equavalent to zip(some_list, some_list[1:]) -- for
generators.
* "consume" consumes an iterator.

As of the date of this commit, there is no way to get an arbitrary
element of an iterable like a set without modifying it (for example, as
in `set.pop()`). The `peek()` function is more readable than writing

    next(iter(S))

whenever an arbitrary element of an iterable is required.

Python 3.2 introduces the `itertools.accumulate()`. Before, we were
using less flexible custom code for performing the same task, in
`networkx.utils.misc.cumulative_sum`. This commit replaces calls to
`cumulative_sum` with calls to `itertools.accumulate()`.

Since we currently still support Python 2.7, this commit adds a fallback
equivalent definition of accumulate that matches the Python 3.2
implementation. Once support for Python 2.7 is dropped, all of this code
can be removed, and calls to `networkx.utils.accumulate()` can be
replaced by calls to `itertools.accumulate()`.

Remove a print statement and some commented out code.