Add CPI to Arm benchmarks

[mkannwischer]

Our target boards allow to obtain the number of exected instructions using DWT which we can use to the cycles/instruction (CPI) which is particuarly interesting for dual-issuing CPUs like the Cortex-M7: https://developer.arm.com/documentation/ka001499/latest/

We should add that to the benchmarks - in particular to the profiling.