Identify and test or store tricky forms to prevent regressions

[lognaturel]

For example: https://code.google.com/archive/p/opendatakit/issues/888

This will be especially important as we're looking for performance improvements. All of the different DAG implementations stem from balancing strict correctness and speed. The more forms there are to actually verify correctness, the better.

@dcbriccetti @mdudzinski

[lognaturel]

I don't think we have any coverage of predicates that look quite like https://forum.opendatakit.org/t/validate-against-duplicate-entries-within-nested-repeats/5840/2

[lognaturel]

I think @ggalmazor has given us a great start in the smoketests package so we can close this issue. Naturally, there'll always be more to do but now there are good examples and a lot more coverage.