Function cuda_batch_rows 

pub fn cuda_batch_rows() -> Option<usize>

Prompt rows per batched-graph chunk when the CUDA backend runs the graphs (CMF_CUDA_BATCH), None on every other backend.