CUDA NVCC compiler lookup should allow build machine #15502
Description
Describe the bug
NVCC is architecture agnostic so it should use the default one even when cross compiling
To Reproduce
Try cross compiling with cuda without defining cuda='nvcc' in the [binaries] section of the cross file
Expected behavior
Can build without specifying the nvcc binary
Seprate issue but might it also be correct to auto set ccbin to the defined c++ compiler?
system parameters
- Cross build x86_64 to aarch64
- Ubuntu 22.04
- Python 3.12.9
- meson 1.9.1
- ninja 1.12.1