Genuine question: Are GitHub workflows stable enough to be used for benchmarking? Like CPU time quantum scheduling is guaranteed to be the same from run to run?