More reliable plugin integration tests

[numirias]

The tests in test/test_plugin.py run against headless nvim instances. Since many features of Semshi operate asynchronously, it can be tricky to track when they have completed (esp. because sometimes we can't differentiate between a pending and a failed event). Consequently, some of the tests are based on timeouts which occasionally makes tests fail randomly on Travis.

This should be fixed soon because it also potentially confuses contributors submitting PRs.

(Also we should revisit the inadequate usage of fixtures that Pytest is nagging about.)