results_parser: Use sorted lists instead of sets for column values

[gruberma]

Following columns contain sets as values:

• verdict columns (Passed_sameOrder, Failed_sameOrder, ...) of the passed-failed overview
• Verdicts_sameOrder, Verdicts_randomOrder of the tests-overview

This is bad, because it makes the resulting CSV output non-deterministic, since the sets have non-deterministic order between different Python executions.
Originally it was used to filter out duplicate result entries (the same test is reported multiple times in the same run), which actually occurs in certain junit-xml files.

Solution: keep the duplicated-dropping set operation, but wrap it in sorted to convert it back to a list.

Watch out for the following also needed adjustments:

• Adjust (and simplify) tests
  • now we can actually assert on string-equality of the output CSV files -> no parsing needed)
• Adjust CSV parsing