As described in issue #33, the option does actually not limit the number of cores but selects specific cores instead. This behavior was not intended

Replace own implementation of parameterized tests with
pytest parametrize

Not supported by Python 3.6

Closes #20

Wrong check was called, so we didn't really check compilation before

The check 'bad-continuation' doesn't exist anymore, so we do not have to disable it.
Similarly, 'no-self-use' was moved to an extension. Because we are no fan of this check
(cf. https://github.com/PyCQA/pylint/issues/5502 ), we do not enable this extension.

Log methods with this attribute use sys.stderr by default

To make testing TestCov easier; no need to require each unti test to initialize
the logger first.

Previously showed one dot per execution, which was difficult to count.

Previously, the first log message was at the end of the same line as the progress display,
and after a log message, the next progress display was just an additional dot.

To make display better readable, change output to the following:
1. The first log message is on a separate line (just as intuitively expected)
2. If a log message is printed, the next progress display is not just
   one dot, but it repeats the full 'Executing tests' with the full number of dots
   that represents all tests executed so far (+1 for the currently executing test)
3. There's no additional empty line between log messages and message "Done!"
   (this was only there if at least one log message was printed, so now
   it's consistent)

When --verbose is used, matplotlib DEBUG messages were displayed before,
because we used the root logger for logging TestCov messages.
To avoid such clutter from subcomponents of TestCov, use a named logger
instead of the root logger.

Closes #29

To avoid me wondering about this code again

In this construct, no __gcov_dump was added before, leading to incomplete coverage measurements:

void reach_error() { __assert_fail(...) }

Do not expect a newline before an abort, or the abort to be indented.