Skipping was done incorrectly and failed on Windows.

This patch replace all uses of nose with pytest. It also includes a
pytest plugin for creating EARL reports for tests with a `rdf_test_uri`
parameter.

Some caveats:

- HTML report directory is now htmlcov instead of coverage
- There is some warning related to the EARL reporting plugin which I can't quite figure out:
  ```
  .venv/lib64/python3.7/site-packages/_pytest/config/__init__.py:676
    /home/iwana/sw/d/github.com/iafork/rdflib/.venv/lib64/python3.7/site-packages/_pytest/config/__init__.py:676: PytestAssertRewriteWarning: Module already imported so cannot be rewritten: test.earl
      self.import_plugin(import_spec)
  ```
  This is not causing any problems as far as I can tell, but still annoying.
- python setup.py test won't work anymore, I can make it work but this
  is not advised by pytest:
  https://github.com/pytest-dev/pytest-runner/#deprecation-notice
- run_test.py is still there but it's not really referenced anymore from
  anywhere and the options it accepts are completely different as it's options
  were based on nose. I would say it should be removed entirely but for now
  it is basically just a wrapper around pytest that basically does nothing.
- Removed references to test attributes as currently they are not being
  used anywhere anyway, I guess we can add them back if there is some
  use for them later.
- A lot of tests are still marked to skip when really they should be marked
  with xfail. This is also affecting the RDFT test manifests and result in
  reports saying tests are skipped when really we know they will fail and
  they are only skipped for this reason. But there is no change here from
  before, and pytest makes it easier to dynamically do expected failures.

Special thanks to Wes Turner for his advice and inputs on this process.

For some reason the sequence of processing is different in Travis than in PyCharm.  Changing the module names eliminates any possibility of confusion.
Also fixed a number of warnings, whines and general unhappiness in the build process.

There is a test that requires both os.fork and os.pipe present. Rather
than enumerating platforms that don't support it, just try to import the
functions and if you can't, skip the test.