build: bump dace version

[edopao]

This pull request includes updates to dependencies, configuration handling for CUDA architecture detection, and a CI job setting. The most significant changes are an upgrade to the required DaCe version, an improved approach to configuring CUDA architecture detection, and a minor CI configuration adjustment.

Dependency and configuration updates:

• Upgraded the required dace package version from 2.0.0a5 to 2.0.0a6 in pyproject.toml, ensuring compatibility with recent DaCe features and fixes.
• Changed how CUDA architecture detection is handled in set_dace_config in common.py: now sets compiler.cuda.cuda_arch directly instead of passing a CMake argument, aligning with best practices and recent DaCe updates.

CI/CD configuration:

• Removed allow_failure: true from the build_cscs_amd_rocm CI job in ci/cscs-ci-dace-determinism.yml, making failures in this job block the pipeline.

[edopao]

cscs-ci run default

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run dace-determinism

[edopao]

cscs-ci run default

[edopao]

@romanc @FlorianDeconinck I see errors in the cartesian tests using the newly released dace v2.0.0-alpha6:
error: use of undeclared identifier '__dace_gpu_drain_error'

That is strange, because the next tests pass. I haven't check yet.

[edopao]

@romanc @FlorianDeconinck I see errors in the cartesian tests using the newly released dace v2.0.0-alpha6: error: use of undeclared identifier '__dace_gpu_drain_error'

That is strange, because the next tests pass. I haven't check yet.

There are errors also in the GH200/CUDA pipeline.

[FlorianDeconinck]

First time we hit this, I have never seen this __dace_gpu_drain_error. We run parts of our model code as part of the DaCe CI to evade this issues, so I am bit suprised

[FlorianDeconinck]

@tobias: so because it's crashing across the board I am guessing our glue code with DaCe for stencil mode has broken down with the new version - any st:dace:gpu:X backend should showcase it

[romanc]

The undefined variable points to https://github.com/spcl/dace/pull/2489/changes. It was introduced there and that's why we've never seen it before.