Function launch_config_with_grid 

pub fn launch_config_with_grid<K: Kernel>(
    grid_x: usize,
    program: &CudaProgram<'_, K>,
) -> CudaLaunchConfig

Build a launch config with a pre-computed grid and block from kernel metadata.

Use when the launch grid (number of CTAs) is known independently of element count — for example, tiled matmul kernels where grid = ceil(M/TILE_M) * ceil(N/TILE_N). The block and cluster dimensions are read from program.metadata.