Only enable example 14 when the backend is CUDA

[FMarno]

Maybe it was already broken, but the recent rebase for CUTLASS 3.6 introduced this line:

https://github.com/codeplaysoftware/cutlass-fork/blame/d49319f72cd9eeeb49830a94e23b3fdf8635fcc8/include/cutlass/gemm/device/gemm_universal_adapter.h#L461

Making use of DispatchPolicy::SubgroupSize. The example examples/14_ampere_tf32_tensorop_gemm/ampere_tf32_tensorop_gemm_cute.cu had set DispatchPolicy to

using DispatchPolicy = cutlass::gemm::MainloopSm80CpAsync<PipelineStages>;

leading to this compiler error:

/tmp/finlay/cutlass-fork/include/cutlass/gemm/device/gemm_universal_adapter.h:461:70: error: no member named 'SubgroupSize' in 'cutlass::gemm::MainloopSm80CpAsync<4>'
  461 |           kernel_properties{sycl_exp::sub_group_size<DispatchPolicy::SubgroupSize>}
      |                                                      ~~~~~~~~~~~~~~~~^

I have modified the cmake to only enable that example for nvidia devices.

[aacostadiaz]

LGTM