Fix CUDA >= 13.0 supported-arch list so DGX Spark (GB10 sm_121a) builds target
the 12.1 family instead of falling back to 12.0.

With CUDA 13.x the fork uses family targets, but CUDA_SUPPORTED_ARCHS for the
CUDA>=13 branch lists only up to "12.0". A GB10 build requested as 12.1a is
therefore dropped by cuda_archs_loose_intersection() down to 12.0, and kernels
using the sm_121a-only FP4 instruction `cvt.e2m1x2.f32` fail in ptxas:

    error: Feature 'cvt.e2m1x2.f32' not supported on .target 'sm_120'

Add the "12.1" (sm_121 family) entry so GB10 compiles for sm_121a.
---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0000000..1111111 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -118,7 +118,7 @@
    # starting from CUDA 12.9 and Blackwell (10.0), we use family-specific targets (10.0f, 12.0f, etc)
    # to support the whole generation without specifying all sub-architectures
    # see: https://developer.nvidia.com/blog/nvidia-blackwell-and-nvidia-cuda-12-9-introduce-family-specific-architecture-features/
-  set(CUDA_SUPPORTED_ARCHS "7.5;8.0;8.6;8.7;8.9;9.0;10.0;11.0;12.0")
+  set(CUDA_SUPPORTED_ARCHS "7.5;8.0;8.6;8.7;8.9;9.0;10.0;11.0;12.0;12.1")
 elseif(DEFINED CMAKE_CUDA_COMPILER_VERSION AND
    CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8)