Add ARM github action, ROCm 7, CUDA 13 gitlab job and corresponding update

[yhmtsai]

It is raised from #1931 and #1867.

they are in quick or additional branch rule in CI setting, but I will revert them before merge

Fixes #1931

For ARM, apart from the benchmark statistic check,
we also have CGS SolvesDenseSystemMixedComplex on complex in https://github.com/ginkgo-project/ginkgo/actions/runs/19371107556/job/55427021354
Currently, using -ffp-contract=off can pass both.

ROCm 7 does not support MI50 officially.
For ROCm 7,
we can not define the warpSize via compiler option because one of their file use the same name as variable.
I only changed that in hip config and checked that all ginkgo based on the value not warpSize.

some extra conditions for kernel because we have grid size = 0 for those kernel. It starts to be capture in ROCm 7

Additional hipGetLastError() in the OverAllocating test because ROCm 7 still has that and be captured by hipGetLastError() in sparse handle creation (so exception be thrown in the next test). We throw it here again to avoid polluting the next test step.
Related Issue: ROCm/rocm-libraries#2801

For custom thrust namespace, ROCm 7.1.0 does not fully adapt the custom namespace in some place, but I believe they fix it in ROCm/rocm-libraries#1769
Need to check it when the new release comes.

For CUDA 13, server currently install it as usr/local package not from spack.
Thus, I add temporary lmod file during pre_script, which should not be necessary with spack setup. I can just keep the corresponding fix without including the ci and lmod. Or, keep the temporary workaround and then remove it when the server has that.

[ginkgo-bot]

Error: The following files need to be formatted:

core/config/config.cpp
include/ginkgo/core/base/combination.hpp
include/ginkgo/core/base/device_matrix_data.hpp
include/ginkgo/core/base/lin_op.hpp
include/ginkgo/core/base/matrix_assembly_data.hpp
include/ginkgo/core/base/matrix_data.hpp
include/ginkgo/core/base/mtx_io.hpp
include/ginkgo/core/base/perturbation.hpp
include/ginkgo/core/factorization/cholesky.hpp
include/ginkgo/core/factorization/factorization.hpp
include/ginkgo/core/factorization/lu.hpp
include/ginkgo/core/factorization/par_ic.hpp
include/ginkgo/core/factorization/par_ict.hpp
include/ginkgo/core/factorization/par_ilu.hpp
include/ginkgo/core/factorization/par_ilut.hpp
include/ginkgo/core/log/batch_logger.hpp
include/ginkgo/core/log/convergence.hpp
include/ginkgo/core/log/papi.hpp
include/ginkgo/core/log/stream.hpp
include/ginkgo/core/matrix/batch_identity.hpp
include/ginkgo/core/matrix/fft.hpp
include/ginkgo/core/matrix/identity.hpp
include/ginkgo/core/matrix/scaled_permutation.hpp
include/ginkgo/core/matrix/sparsity_csr.hpp
include/ginkgo/core/multigrid/fixed_coarsening.hpp
include/ginkgo/core/preconditioner/batch_jacobi.hpp
include/ginkgo/core/preconditioner/gauss_seidel.hpp
include/ginkgo/core/preconditioner/isai.hpp
include/ginkgo/core/preconditioner/sor.hpp
include/ginkgo/core/reorder/mc64.hpp
include/ginkgo/core/reorder/nested_dissection.hpp
include/ginkgo/core/reorder/rcm.hpp
include/ginkgo/core/reorder/scaled_reordered.hpp
include/ginkgo/core/solver/batch_bicgstab.hpp
include/ginkgo/core/solver/batch_cg.hpp
include/ginkgo/core/solver/bicg.hpp
include/ginkgo/core/solver/bicgstab.hpp
include/ginkgo/core/solver/cb_gmres.hpp
include/ginkgo/core/solver/cg.hpp
include/ginkgo/core/solver/cgs.hpp
include/ginkgo/core/solver/direct.hpp
include/ginkgo/core/solver/fcg.hpp
include/ginkgo/core/solver/gcr.hpp
include/ginkgo/core/solver/gmres.hpp
include/ginkgo/core/solver/idr.hpp
include/ginkgo/core/solver/ir.hpp
include/ginkgo/core/solver/triangular.hpp

You can find a formatting patch under Artifacts here or run format! if you have write access to Ginkgo

[MarcelKoch]

Shouldn't we also add this to ginkgo_cuda?

[MarcelKoch]

Suggested change

- 'arm_action'

I think this caused the CI to run twice.