Revert "cuda: add v13.1.2, v13.2.0, v13.2.1 and v13.3.0"

[sethrj]

Reverts #4897 : downstream packages including strumpack seem to fail with it; they weren't tested on gitlab CI due to an error introduced by #4615 and fixed in #5046.

Draft to test whether the pipeline passes with the reversion.

[rbberger]

@sethrj instead of reverting, you should document which ones are failing for you and add conflicts.

[sethrj]

This unfortunately is due to a lapse in the CI system; it's the responsibility of each PR to make sure that new capabilities don't break other code. I'm mostly trying to verify that the CI works after this; a couple of software packages have been broken on develop since #4897 .

https://gitlab.spack.io/spack/spack-packages/-/pipelines/1623929

[alalazo]

@sethrj Use the don't merge yet label and make this a normal PR. Pipelines don't run on draft PRs 😬

[balay]

https://gitlab.spack.io/spack/spack-packages/-/jobs/22451976

/tmp/root/spack-stage/spack-stage-petsc-3.25.1-lxmyy25uscljmm3jlejoth6zcia3lveg/spack-src/src/mat/impls/diagonal/cupm/cuda/../matdiagonalcupm.hpp(43): error: namespace "thrust" has no member "tuple"
    using argument_type = thrust::tuple<PetscScalar, PetscScalar, PetscScalar>;

FWIW cuda@13.3 build issue is fixed in petsc@3.25.2 - that got similarly merged to develop via

#5053

[balay]

@sethrj Use the don't merge yet label and make this a normal PR. Pipelines don't run on draft PRs 😬

The label can be easily ignored [I've had that issue once].

Would be nice if the restriction on draft PRs can be removed [or be able to add in CI via a lablel - or other mechanisms]

[balay]

Is it worth checking if just reverting cuda-13.3 (or making cuda-13.2.1 the default) would work - without a full revert?

[balay]

And I see tau, strumpack have PRs related to cuda-13.2+ merged in (along with petsc).

Likely outstanding pkgs [from the above pipeline failure]:

• visit
• omega-h

[balay]

And both visit and omega-h failures appear unrelated to cuda-13.2

• https://gitlab.spack.io/spack/spack-packages/-/jobs/22456141/viewer
• https://gitlab.spack.io/spack/spack-packages/-/jobs/22452092/viewer

[rbberger]

Revert seems overkill at this point.