Better benchmarks

[magnus-nomono]

Right now we only run one iteration for each test in the benchmarks. We have, especially for the small array tests, quite a bit of variation in the execution time. We should do more proper benchmarks, preferably utilizing some kind of benchmark package like pytest_benchmark