B200 GPU Support

[lclemens05]

Autotuned parameter file for NVIDIA B200 Blackwell GPU

parameters_B200.json

[alazzaro]

@lclemens05 thanks for your contribution!
could you add B200 to the cmake list of support architecture and make sure they are selected during the build?

[DCM-Uni-Paderborn]

Thanks for adding the tuned B200 parameter set. I tested the current PR head on an NVIDIA B200 node with CUDA 12.8 while preparing a CUDA-enabled CP2K/Skala build.

Two CMake changes are still needed:

1. Please add set(GPU_ARCH_NUMBER_B200 100). With the current head, configuration prints

   GPU target architecture: B200
   Kernel parameters: B200
   GPU architecture number:
   

   so the CUDA_ARCHITECTURES target property remains empty.
2. Please keep the existing P100 default and its help text. Selecting B200 should be explicit; changing the global default would alter existing builds unexpectedly.

It would also be useful to add "parameters_B200.json": "sm_100" to the gpu_architectures mapping in src/acc/libsmm_acc/kernels/smm_acc.py so the tooling recognizes the new parameter file consistently.

For comparison, a local DBCSR build with B200 -> 100 completed successfully and was linked into CP2K. Native-grid Skala smoke tests then passed on one and four B200 GPUs, including deterministic rank-to-device mapping 0:0 1:1 2:2 3:3 and CPU/GPU agreement for energy, forces, and stress.

One independent CUDA 12.8 issue remains: cuFuncSetSharedMemConfig is deprecated and DBCSR promotes that warning to an error. I suggest handling that separately rather than suppressing all deprecation diagnostics in this architecture PR.