faster failfast by cancelling all test-items running in parallel when one fails

[nickrobinson251]

When running with nworkers > 1 there may be multiple test-items running when one fails. Currently, when failfast=true , a failure prevents any further testitems starting, but will wait for the currently running ones to complete... which might be as long as timeout * nretries...

Instead, when one test-item fails we should proactively interrupt the other test-items (and kill the workers, and side-step any retry-logic, and log that these tests were interrupted/cancelled). This would allow failfast to report the failure and stop the tests much faster

We need to decide how the testitems that are interrupted (because a different testitem failed) should appear in the printed test summary (and the xml test report) -- perhaps they should appear as skipped? Or simply omit them? Omitting them might be confusing since it'd be inconsistent with the logs, but could still be an okay option... we can't mark them passed or failed (because they didn't complete), so omitting them or marking them skipped are i think our only options.

Follow-up to #133