All QA tests expect to be able to log in at the start of the test.

That's not possible if a test leaves the browser logged in when it
finishes. Normally this isn't a problem because Capybara resets the
session after each test. But it does that in an `after` block, so when
a test, like this one, logs in in an `after(:all) block, the browser
returns to a logged in state *after* Capybara has logged it out. And
then the next test will fail.

E.g.: https://gitlab.com/gitlab-org/gitlab/issues/34736

This adds a `config.after(:context)` block to fail if a test
leaves the browser logged in.