Faster implementation of WakePotential class and binning of particles

Hey,

I've implemented a new version for Wakepotential class that makes tracking call to it x5-10 times faster (from 150-300 ms/call on my laptop to 30 ms/call). The overall simulation is 1.5-2 times faster. All tests have passed and I've also rerun the benchmark (which took 3 hours) of Head-tail instability on my laptop (picture below). I had to assume that the growth rates in your benchmark are not true growth rates of the instability but the growth rates of Courant-Snyder invariant (so, 2x slower) to get a match. Otherwise, the agreement is perfect. Let's check this merge request together. I will also make some minor adjustments. Also, we can remove interp_on_position option because the code is now very fast.