Complete CPU run: all 16 models on all three devices, and the throughput finding

Add CPU to the device comparison and a generated evaluation figure across CUDA, CPU and MPS