Debian Trixie and Sid fail to build the CUDA kernels with clang

[totaam]

We have to install clang++ to build the CUDA kernels because of #4281

But now we hit a different issue:

Error: failed to compile CUDA kernel BGRA_to_RGBAP
 using command: ['/opt/cuda-12.6.2/bin/nvcc', '-fatbin', '-std=c++11', '--allow-unsupported-compiler', '-ccbin=clang++', '-arch=all', '-Wno-deprecated-gpu-targets', '-Xnvlink', '-ignore-host-info', '-c', 'fs/share/xpra/cuda/BGRX_to_YUV444.cu', '-o', 'fs/share/xpra/cuda/BGRX_to_YUV444.fatbin']

/usr/lib/gcc/x86_64-linux-gnu/14/../../../../include/c++/14/type_traits(1610): error: "__is_nothrow_new_constructible" is not a function or static data member
                        constexpr bool __is_nothrow_new_constructible
                                       ^
/usr/lib/gcc/x86_64-linux-gnu/14/../../../../include/c++/14/type_traits(3066): error: type name is not allowed
          bool _Dangle = __reference_converts_from_temporary(_Tp, _Res_t)
                                                                  ^

I am not going to delay the 5.x release for this.