[clang] [CUDA] Add device-side kernel launch support (PR #165519)

[via cfe-commits]

darkbuck wrote:

I should mention that the generated code sequence differs from the CUDA 12.9. The latter uses `cudaGetParameterBufferV2` and `cudaLaunchDeviceV2` interfaces, which are documented in the CUDA programming guide. I verified the device-side kernel launch with the dynamic mandelbrot app (https://github.com/darkbuck/mandelbrot-dyn.git). It works with CUDA SDK 12.9 on my RTX 4060.

https://github.com/llvm/llvm-project/pull/165519