Failures will not longjmp, but are reported on top of the stack,
by calling the case failure handler, which then decides whether to
continue or abort testing.
Other changes:
- Individual cases can have their own setup and teardown handlers,
- Case repeats are now handled by the case handler returning `control_flow_t`,
- Separates `status_t` and `failure_t`,
- Implements useful default case setup and teardown, and
- Keeps score of passes and failures on per case and test basis.