Remove Python 2 support

[groutr]

@eriknw, this PR removes Python 2.7 support.

[groutr]

@eriknw this is ready for review.

[groutr]

@eriknw This should be ready now.

[groutr]

Added Python 3.9-dev to tested versions.

[groutr]

@eriknw I restored the Python 3 section of compatibility.py and added a deprecation warning. This feels like a pretty cluttered PR (my travis ci config skills are not great), perhaps I should clean up the commits and open a new PR.

[groutr]

I need to confirm, since I think I crossed my wires in my head, should toolz continue to advertise support for Python 3.4 if we don't test against it?

[eriknw]

Thanks again for all this @groutr . I'm fine dropping (and not advertising) Python 3.4 since it's already end-of-life. Python 3.5 only has six more months.

[eriknw]

Looks great, let's get this in today!

[eriknw]

Oh, and .coveragerc no longer needs the pragma: py$MAJOR_PYTHON_VERSION no cover trick.

[groutr]

@eriknw Finished removing references to Python 3.4 and fixed the pragma checks.

[eriknw]

LGTM. Shall we merge? Thanks again for doing the heavy lifting and being patient and persistent.

[groutr]

I think this is in a merge-able state. There are some more things that I will do in another PR that will build off this.
Namely:

1. In itertoolz.dissoc, use dict views instead of building a set. IIRC, this using views was a faster than creating the set (the set was required to make the code Python 2 compatible).
2. Use yield from in places where it makes sense. This can provide a small performance increase in the cases where there are loops like

for item in seq
    yield item