Make it easier to create backwards compat tests and test across a wider array of pipeline types.

1. Test correctness against a medium-sized corpus of different pipeline types. Correctness evaluates both:
   1. Correct output
   2. Correct stats (especially input/output message counts)
2. The test framework should facilitate the below for each pipeline/input:

1. 1. Test results are correct and equivalent even after checkpoint restore.
   2. Automated creation of checkpoint files for backcompat testing.
   3. Validate correct results when restoring from old format checkpoint data (previously generated using 2 above)
2. Easy to manually inspect that the expected output in each test case is correct.