Enable NVTX in cuvs-cagra-search component (#439)

Since parts of CAGRA code have been separated into a static library component `cuvs-cagra-search` (to selectively enable CUDA separable compilation on them), the NVTX flags are not passed to the affected sources anymore. This PR fixes that.

Authors:
  - Artem M. Chirkin (https://github.com/achirkin)

Approvers:
  - Micka (https://github.com/lowener)

URL: #439

cpp/CMakeLists.txt

@@ -603,6 +603,9 @@ SECTIONS
     # This enables NVTX within the project with no option to disable it downstream.
     target_link_libraries(cuvs PUBLIC CUDA::nvtx3)
     target_compile_definitions(cuvs PUBLIC NVTX_ENABLED)
+
+    target_link_libraries(cuvs-cagra-search PUBLIC CUDA::nvtx3)
+    target_compile_definitions(cuvs-cagra-search PUBLIC NVTX_ENABLED)
   else()
     # Allow enable NVTX downstream if not set here. This creates a new option at build/install time,
     # which is set by default to OFF, but can be enabled in the dependent project.