Slow jitting benchmarks

[pnkraemer]

#14 introduces a benchmark-option that just-in-time-compiles the attempt_step functions.
This is for two reasons:

1. It is interesting to see how much (wall-clock) run time one can gain by using a few additional tricks
2. The run-time curves will look less noisy ("cleaner").

Since this makes the experiment really slow (compilation takes a while), it is turned off by default and should only be turned on when desired.

If we notice that we never turn it on, let's remove it again!

[pnkraemer]

At the moment the jitted experiments look better than the non-jitted:

1_attempt_step_lorenz_timing.pdf

(better in the sense that growth rates are more evident). Maybe that is an argument in favour of keeping them...!? Let's just keep an eye on them

[nathanaelbosch]

Without having experienced the slow compile times, I'm strongly in favor of just jitting everything we can! It seems like a super simple way to make things faster and so I don't see a good reason for reporting the slower runtimes, if we also have faster ones.

And the curves really do look much nicer :)