More benchmark machine args

[ggwpez]

I ran the benchmark machine command some dozen times on the ref hardware and the Blake2-256 seems to be the only failing one. It runs for only 100ms which is probably too short to get reproducible results.

Changes:

• Increase benchmark durations
• Add arguments to control all durations
• Let sc-sysinfo return f64 for more precision
• Add explicit default limits to sc-sysinfo

The benchmark worked 27 times in a row, then failed with 0.4% outside the tolerance for the disk benchmark. Should be good enough for now.

[koute]

So what's the probability of it failing with only 100ms time? How does it change if you'd increase it to 200ms? (I'm thinking that we might consider increasing the benchmark time slightly for telemetry too, although when I was testing it out on the benchmarking machine it seemed stable enough.)

[ggwpez]

@koute yea good idea.
So with 100ms it often fails with down to just 73%. With 200ms the lowest that I am seeing is 85%. With 400ms it only dips to 92% and passes fine.
Not sure if the benchmarks influence each other or what is going on, but sometimes I can consistently reproduce these numbers and when I try it the next time; not at all.
Maybe we need some dedicated test for this, but for now I dont have the time.

[ggwpez]

bot merge