NEMO Reproducibility issues

[sergisiso]

We can now run NEMOv5 with full numerical reproducibility if conservative flags are used. However, this is not the case for NEMOv4. We should identify the exact file-routine-loop that causes these divergences to exclude them and track the issues.

[sergisiso]

Also, I have found that different compilers with different behaviour (also for NEMOv5), so I think we will need to test them all (at least for CPUs), which will increase the integration test time. But to compensate maybe could make the passthrough step an optional action because the threaded transformation already contains the passthrough check implicitly. Then when a threaded compiler check fails we would have the possibility to manually trigger the passthrough action to see if it was a frontend issue or a transformation issue. What do you think @arporter ?

[arporter]

Having the option to trigger different tests only when required sounds good to me. We could potentially have them run on a different machine too.