This PR fixes ECPAIR glue in case when the last input pair is not non-trivial (either (0,0) or (0, Q) or (P, 0)). Previously, two things happened when the last input was half-trivial (only G2 membership check i.e. (0, Q)) - it was sent for the G2 membership circuit for testing (as requested by the arithmetization) and we didn't mark the last non-trivial input to go to the ML+finalexp circuit (as it was decided by the input pair ID relative to the total number of pairs).

To overcome that, we now in the glue index the input pairs differently. Instead of following the input pair IDs coming from the arithmetization, we count it ourselves only for the inputs which go to the pairing circuits. And for the "total pairs" we count only non-trivial pairs (as the ML of trivial pairs is 1 and doesn't change the accumulator. What only matters is that they are valid i.e. belonging to the correct subgroups).

To make it work, we had to add a constraint to ensure the total number of pairs was correctly set. But the constraints to ensure sequential counting of the non-trivial pairs still follows from the previous constraints (we check that after every 60 rows the input ID increases).

Additionally, we had to ensure the accumulator consistency correctness. When we have only one valid pair, then we only call ML+finalexp circuit. In this case we cannot check the consistency of the accumulator, but instead have to check that the previous accumulator going to the ML+finalexp circuit is correctly initialized.

Another fix I had to implement is to allow to have pairing circuit inputs and G2 membership inputs in the same rows (see 60ee51c). Previously when a row was an input to the G2 membership circuit, then the index constraint was not enforce, but should have been.

Finally - added regression test for a transaction which incurred the bug.

Also did some cosmetic changes:

• generated testdata modules for some test cases
• unified the testdata modules to have same order of columns
• added option InHex to csvTraces.FmtCSV method to force dumping also small outputs in hex. This is mainly to have minimal diff of the current testcase modules.
• defined a utility method for dumping the testdata module. It is useful for creating more regression tests.

On top of that, I added a comprehensive test case generator which generates all possible 1-5 input pair combinations for the pairing check. Currently the tests are skipped as it generates about 65000 test cases, but can be manually called.

Checklist

• I wrote new tests for my new core changes.
• I have successfully ran tests, style checker and build against my new changes locally.
• I have informed the team of any breaking changes if there are any.