run-aa-test: report the underlying tests and the log parsing separately

Before this change, if "normal" failed due to a bug in the underlying
tests and "malicious" failed due to an unmet expectation, the
machine-readable parts of our log would be

normal.expected: fail
malicious.expected: fail

and discovering the reasons would require reading logs. Now, we would
log that situation as:

normal.expected_underlying_tests: fail
normal.expected: pass
malicious.expected_underlying_tests: pass
malicious.expected: fail

and an appropriate developer can investigate in the right places; in
this case, the "normal" failure would require someone who knows
about whatever is under test, for example Tracker, while the
"malicious" failure would require someone who knows about AppArmor.

Differential Revision: https://phabricator.apertis.org/D277


Signed-off-by: Simon McVittie <simon.mcvittie@collabora.co.uk>
Reviewed-by: xclaesse