Add perf:int for arithmetics.k using the same in-process structure as perf:ieee.

Make perf output describe the benchmark lanes and report per-iteration samples with averages.

Also, enforce uniqness of --lib option