in matmul naive vs tiled demo, now using training kernels that produce...

in matmul naive vs tiled demo, now using training kernels that produce access-to-footprint-ratios that more closely match test kernels