Support Python 3

[ghost]

Part of #232

Cola now runs on both Python 2.6+ and 3! Many refactorings were
necessary for this. The first commits are about modernizing the py2 codebase and adding necessary __future__ imports, and the last commit is the real thing.

In-house "six-like" compat unit: Because there were only a few
compatibility tricks needed, I felt that it was overkill to add a
dependency on the "six" module, so I added the necessary compatibility
tricks in cola.compat:

• ustr: replaces all previous instances of unicode.
• unichr: Alias of chr under py3.
• StringIO: Import from cStringIO under py2 and io under py3.
• urllib: Aliases to urllib.parse under py3.

Then, there's a couple of small fixes that had to be made:

• StandardError --> Exception
• except foo, e --> except foo as e
• foo.reverse() --> reversed(foo)
• xrange() --> range() (minimal perf impact under py2)
• iteritems() --> items() (minimal perf impact under py2)
• Wrapping list() where needed (filter(), map(), etc.)
• __next__ = next aliasing

Also, there were a couple of places where core.encode() was used and
didn't have to. Under py2, it did nothing, but under py3, it generated
an error. I've tried to be as careful and minimal as possible for that
part, but if there's one area where there's a risk of newly indtroduced
bugs, it's that part.

There were also "cmp" functions that had to be replaced with "key"
functions.

[davvid]

Looks great!I think once we fixup these two small issues it'll be good to go. Yeah, StringIO instead of cStringIO would be a fine solution. It definitely has saner behavior.

>>> from StringIO import StringIO
>>> io = StringIO()
>>> io.write(unichr(0x1234))
>>> io.getvalue()
u'\u1234'

[ghost]

Well, it turns out that Python added its io module in v2.6. I didn't know that. That means that we don't need a compat.StringIO and we can simply use io.StringIO everywhere.

I've tested the new code with non-ascii data under both v2.7 and v3.3 and it worked fine.

[davvid]

Awesome, thank you!