Fix CUDSS extension for CUDA v6.x

[albertomercurio]

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

Given the recent updates of CUDA.jl and CUDSS.jl, we need these small changes in the extension here.

[ytdHuang]

By the way, we also need a version release after this PR.

[ChrisRackauckas]

Test with CUDSS loaded: Test Failed at /home/chrisrackauckas/github-runners/demeter4-2/_work/LinearSolve.jl/LinearSolve.jl/test/gpu/cuda.jl:51
  Expression: LinearSolve.cudss_loaded(A_gpu_csr)

interesting, it didn't load?

[ytdHuang]

@albertomercurio @ChrisRackauckas

It seems like the GPU test is still using CUDA v5.11.2 but with CUDSS v0.7.0.

So when we call :

using CUDA.CUSPARSE # from CUDA v5

using CUDSS
CUDSS.cuSPARSE # from cuSPARSE v6

it is indeed different I think

[ytdHuang]

By the way, would it be better that the Project.toml is setup as:

[extensions]
LinearSolveCUDAExt = ["CUDACore", "cuSPARSE"]

instead of "CUDA" ?

[ChrisRackauckas]

Possibly, I haven't looked too much at the recent changes.

[ytdHuang]

I tried it locally, but the compat version cannot be resolved.

Because CUSOLVERRF.jl only support until CUDA v5.

[ChrisRackauckas]

So we need to drop or update that? Is CUSOLVERRF.jl still maintained?

[ytdHuang]

I opened an issue in CUSOLVERRF today: exanauts/CUSOLVERRF.jl#27

But the last version release for CUSOLVERRF was Apr. 18, 2025.

===

I think the best solution for now is to drop CUSOLVERRF, and use cuSOLVER directly instead of CUDA.

===

I tried to fix all CUDA v6/CUDSS v0.7 related stuffs in PR #986