Fix set_level side effects

[chekunkov]

On test case completion all modifications to loggers levels are reversed.

Fixes #60

[chekunkov]

Have no idea why pypy3 tests fail, they are broken in master.

Regards 2.6 tests failure - I have two options:

1. Have if statement with some fix only for Py 2.6.
2. Drop .Py 2.6 support.

I'd go with the latter option, it seems to be trending among major Python packages:

https://github.com/pypa/pip/issues/3955 https://github.com/pypa/setuptools/issues/878 https://github.com/pytest-dev/pytest/pull/2031

WDYT?

[hugovk]

It's time to drop 2.6, it's been EOL since 2013:

• https://en.wikipedia.org/wiki/CPython#Version_history
• https://snarky.ca/stop-using-python-2-6/
• http://www.curiousefficiency.org/posts/2015/04/stop-supporting-python26.html
• http://www.python3statement.org
• Current pip 9 deprecates Python 2.6 support, pip 10 won't support it (https://github.com/pypa/pip/issues/3955)
• Not much PyPI traffic (June 2016) https://github.com/pypa/pip/issues/3796

[hugovk]

Here's the pip installs for pytest-catchlog from PyPI for the last month (via pypinfo --percent --pip pytest-catchlog pyversion)

python_version	percent	download_count
2.7	57.1%	25,939
3.6	18.4%	8,337
3.5	14.4%	6,535
3.4	6.6%	3,014
3.3	2.6%	1,194
3.7	0.8%	363
2.6	0.1%	25

I think it's safe to drop 2.6!