[BUG] Build failure when compiling cuvs-25.08 from source: CUDA sync primitives require sm_70+ architecture

Describe the bug
When building cuvs-25.08 from source, the following error occurs after running make -j4:

error "CUDA synchronization primitives are only supported for sm_70 and up."

The compilation will succeeds, if I manually add set(CMAKE_CUDA_ARCHITECTURES "70;75;80;86" CACHE STRING "CUDA architectures to build for") to cpp/CMakeLists.txt :

if(NOT BUILD_CPU_ONLY)
include(rapids-cuda)
rapids_cuda_init_architectures(CUVS)
set(CMAKE_CUDA_ARCHITECTURES "70;75;80;86" CACHE STRING "CUDA architectures to build for")
list(APPEND lang_list "CUDA")
endif()

Steps/Code to reproduce bug
I followed the guide of https://docs.rapids.ai/api/cuvs/legacy/build/#build-from-source and encountered the error:

conda env create --name cuvs -f conda/environments/all_cuda-129_arch-x86_64.yaml
conda activate cuvs
mkdir build;cd build;
cmake -D BUILD_TESTS=ON -DCMAKE_INSTALL_PREFIX=$CONDA_PREFIX ../
make -j4

Expected behavior
The Official guide should lead to a successful compile.

Environment details (please complete the following information):

• Environment location: Bare-metal
• Method of cuvs install: from source
• My nvidia driver version: 575.64.03
• My cuda versiton: 12.9
• My os: ubuntu-22.04 lts

Additional context

Start with cpp/CMakeLists.txt, especially the rapids_cuda_init_architectures(CUVS) call and the documented CMAKE_CUDA_ARCHITECTURES workaround. Reproduce the failure using the listed conda, cmake, and make commands, then verify that a source build completes successfully without the manual workaround.