Fix cudacpp getGoodHel to define a good helicity as ME>0 (fix mismatch in cross sections for fortran and cpp in gg_uu)

[valassi]

WIP: add gu_ttu (with different cross sections for fortran and cpp)

This is a MR to investigate #630. The issue does not only affect gq_ttq where q is anyone of 8 quarks/antiquarks, it also affects simply the u quark.

[valassi]

This MR is now ready for merging.

The MR contains a lot of investigations about comparisons of fortran and cpp cross sections in tmad tests for several processes, including gq_ttq but especially focusing on gg_uu. Many of these investigations will result in opening new issues shortly.

Of the main issues that was identified, specifically for gguu, following up on work by @zeniheisser, is that the old getGoodHel implementation was wrong. The goal was to define a good helicity as an helicity provided a non-zero ME contribution, but the implementation was unable to detect some helicities contributing a relative contribution lower than 1E-16 (or whatever the FP precision is). Therefore gguu in cudacpp was using 6 instead of 8 good helicities, resulting in different cross sections from fortran. This is now fixed and the contribution is compared to 0 (eventually, maybe the same LIMHEL will be used in fortran and cudacpp).

The main benefit of this MR is therefore the fix for getgoodhel (this closes #630). In the investigation I had added gguu.mad to the repo, but I am now removing it to avoid having too many processes there.

Note that gq_ttq cross sections are still wrong, this is now moved to a new issue #748.

[valassi]

The CI tests have passed, I am self merging this one.