Add CMake option to control CUDA host compiler

This revision allows to specify CUDA host compiler (nvcc's -ccbin command
line option) when configuring the build. It addresses the case where the
C/C++ compiler to be used in CUDA toolchain should be different from the
default C/C++ compiler, for instance in case of compilers versions conflicts
or multiple installed compilers.

The new CMake option is named `CUDA_HOST_COMPILER` and can be used as follows:
`cmake -DCUDA_HOST_COMPILER=<path-to-host-compiler>`

If the option is not specified, the build configuration behaves as previously.

Differential Revision: https://developer.blender.org/D14248

intern/cycles/kernel/CMakeLists.txt

@@ -407,6 +407,11 @@ if(WITH_CYCLES_CUDA_BINARIES)
       -o ${CMAKE_CURRENT_BINARY_DIR}/${cuda_file}
       -Wno-deprecated-gpu-targets)
 
+    if(CUDA_HOST_COMPILER)
+      set(cuda_flags ${cuda_flags}
+        -ccbin="${CUDA_HOST_COMPILER}")
+    endif()
+
     if(WITH_NANOVDB)
       set(cuda_flags ${cuda_flags}
         -D WITH_NANOVDB