[CUDA] Support NVTX in CUDA 13 (#18300)

This PR adds the support of NVTX for CUDA 13. The change is because
that starting CUDA 13, the nvtx functions are moved to the lirbary
of `libnvtx3interop.so`, and the previous nvToolsExt library no
longer exists.

To ensure compatibility with both CUDA 12 and 13, we add
libnvtx3interop.so to the library lookup list.
diff --git a/cmake/utils/FindCUDA.cmake b/cmake/utils/FindCUDA.cmake
index 2036c7c..c4c18ee 100644
--- a/cmake/utils/FindCUDA.cmake
+++ b/cmake/utils/FindCUDA.cmake
@@ -101,7 +101,7 @@
         PATH_SUFFIXES lib lib64 targets/x86_64-linux/lib targets/x86_64-linux/lib/stubs lib64/stubs lib/x86_64-linux-gnu
         NO_DEFAULT_PATH)
       find_library(CUDA_NVTX_LIBRARY
-        NAMES nvToolsExt nvTools nvtoolsext nvtools nvtx NVTX
+        NAMES nvToolsExt nvTools nvtoolsext nvtools nvtx NVTX nvtx3interop
         PATHS "${CUDA_CUDART_LIBRARY_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}" ENV LD_LIBRARY_PATH
         PATH_SUFFIXES "lib64" "common/lib64" "common/lib" "lib"
         DOC "Location of the CUDA Toolkit Extension (NVTX) library"