fix: resolve CUDA 12.6 compatibility issue with clang-18

- Use --cuda-path=/usr/local/cuda instead of manual include paths
- This allows clang to properly handle CUDA headers and avoid conflicts
- Fixes '__noinline__' macro conflicts between CUDA and C++13 standard library
- Simplifies build configuration and improves cross-platform compatibility

Author: dr-xjx

attach/nv_attach_impl/trampoline/Makefile

@@ -14,7 +14,7 @@ all: ../trampoline_ptx.h test
 	@echo "Generated $@"
 
 default_trampoline-cuda-nvptx64-nvidia-cuda-$(PTX_TARGET_ARCH).s: default_trampoline.cu
-	$(CLANG) -S $< -Wall --cuda-gpu-arch=$(PTX_TARGET_ARCH) -O2 -stdlib=libstdc++ -I/usr/include/c++/13 -I/usr/include/x86_64-linux-gnu/c++/13 -I/usr/include/c++/12 -I/usr/include/x86_64-linux-gnu/c++/12 $(CUDA_LIBS)
+	$(CLANG) -S $< -Wall --cuda-gpu-arch=$(PTX_TARGET_ARCH) -O2 --cuda-path=/usr/local/cuda $(CUDA_LIBS)
 
 # compile it to exec for test
 test: default_trampoline.cu