Make CenterIdx in VoronoiTessellator consistent/seed-dependent

[lamoreel]

@ilateur : Should be consistent, if ever we wanted to implement a moving frame.

How though? user defined maxRegion, and otherwise an educated guess?

After some thought, I would suggest to simply take a region between 0 and maxRegion. The educated guess can be an extra python function, which the user can call themselves. For example a function to guess the total number of regions that would be generated, or a function that guesses how many regions you need in order to almost never have a "region collision" (I'm guessing this is is huge though).

[lamoreel]

I've done the math; in order to have a 95% chance of never having a "region collision", you'd need at least (approx) 156 region indices (i.e. draw numbers from a uniform distribution with an upper limit of 156). This seems low enough as a default value for maxRegion and is also an overestimate.

(For a 99% chance, you'd need 796 region indices...).

In case of a 3D tessellation, you'd need 273 as maxRegion (for a 95% chance of no collision).

[lamoreel]

This number is also reduced when grainsize is large in relation to the size of the grid. This means that mumax3's upper limit of 256 regions is a (very safe) overestimate (something I wasn't sure about before).
Now the question becomes, what do we want? Our interParameter buffers are filled "bottom-up" starting at zero. This means we want as less regions as possible. How much is physics altered when we set a default upper limit on region indices at 156 (the number of regions can of course be much larger, but regions are assured to be uncorrelated, I think?).

[ilateur]

What kind of magnet length to grainsize ratio did you use to get to this number? 🤔
Also 156 is a gross number, 256 is right there! (or 128?)

But good to hear that an upper limit is probably physically fine. I still don't like bottom-up-filling due to the history dependency, so I would take some "random" number modulo maxRegions? Although this leaves a lot of "holes" in InterParameter

[lamoreel]

The number 156 comes from the fact that a certain region typically has no more than 6-7 neighboring regions (2D). The upper limit of 156 of a uniform integer distribution assures that the chance of 2 neighboring regions having the same region index is less than 5%. If the grainsize is relatively big, this typical number of neighbors decreases, leading to a smaller upper limit.

I don't want to use a modulo operator because this is very susceptible for correlations between different region indices, screwing with the (pseudo) randomness.