We need to detect when a task hasn't finished properly. It's impossible
to do for STI, but for Taskotron generic tasks, we can check whether
results.yml exists. If it doesn't, the task clearly hasn't finished
successfuly. In that case, exit with non-zero exit code, so that our
tools (Buildbot) can inform us about a crashing task.

This code was written after all executions of task-rpmdeplint were
silently crashing every single time for a longer period of time. This
should help us detect such cases.

This will also detect cases when our playbooks are not working properly,
or when task playbooks are not syntactically correct.

Documentation was adjusted to explain that results.yml file must be
created, and how to create that file if you want no results to be
posted.