Update 2D simple shear CI tests

[adigitoleo]

Closes #110.

Uses symmetry diagnostic instead of expensive strength diagnostics which still require further clarification.

Need to convince ourselves that the remaining differences from Kaminski 2001 and 2004 are justified:

• For the GBM test, max. difference is about 5.7 degrees, we don't reach 0 degrees at the 200% strain cutoff
• For the GBS test, the parameter is mostly influencing point-symmetry but only slightly slowing down rotation of the mean a-axis, not by 10+ degrees after 200% strain. Our a-axis also rotates immediately without the initial stagnation (there is no theoretical grounds for the initial stagnation I believe).

[adigitoleo]

Quirky IndexError reproduces on cosgrove (ubuntu, Python 3.11.3) but not for me (Arch and Void, 3.11.4). Can also reproduce, something about the pytest CLI flags, it wants the --outdir?.

[Patol75]

Getting there! Let me have a look at the CI failure.

[Patol75]

Current CI reports two minimal rtol violations and two memory allocation errors. For the latter, they are likely related to the number of grains increasing from 3375 to 4394. Either decrease it or use float32 for that specific test; what do you think?

[adigitoleo]

Current CI reports two minimal rtol violations and two memory allocation errors. For the latter, they are likely related to the number of grains increasing from 3375 to 4394. Either decrease it or use float32 for that specific test; what do you think?

It's one of each, the pytest "capture" output makes errors look duplicated. The number of grains matches the value in the GBS paper (Kaminski & Ribe, 2004, section 4.1 second line) but since there are bigger issues at the moment with matching the GBS results maybe decreasing is fine.

[adigitoleo]

Either decrease it or use float32 for that specific test; what do you think?

Because the tests are now converted to ensemble tests that take a long time, I have disabled them by default and they must be enabled by passing the --runslow flag to pytest, so for now I'm leaving the number of grains as is (running the ensemble has higher memory requirements anyway).