Prioritize kernels for layers using all processors

[rotx-eva]

trafficstars

…ity of "holes" in the kernel map

[oguzhanbsolak]

I did the regression test. I observed no side-effects; the only difference was the prioritized placement of layers that utilize all processors as expected.

It is worth mentioning that latency was affected negatively for ai85-cifar-100-residual and ai85-resnet experiments. The modification increases layer 12's PreRead cycles. The layer 12 depends on the layer 9's and layer 11's output. Both of these layers are placed in larger offsets with the new modification which might be the reason for this latency increase.

I attached ai85-cifar-100-residual log files below. latency_new.txt log_cifar_new.txt log_cifar.txt latency.txt

Other than these experiments, I observed no increase/decrease in the latencies, so I think this is a minor effect and PR is good to go.