Give beeswarm plots fixed limits given input data

[jkrumbiegel]

Currently, whenever you call autolimits! on an axis with a beeswarm plot in it, you get a different result, because it takes the limits of the current scatter plot, but the algorithm finds a new result given those new limits.

I think it makes more sense to pick fixed limits and then adjust axis size or marker size if that's not enough space. You don't get "more" space for the beeswarm anyway by expanding limits, as the extent depends only on screen space.

This PR simply uses the heuristic that having the same space available for each swarm looks good, so it sets data limits that way. If swarms extend out of an Axis, that probably means that they also overlap each other in the middle anyway. So you're unlikely to get a better image by deviating from the heuristic.

[asinghvi17]

Thanks, this looks good! One small change - if there is a gutter then we already have the minimum separation, so that should be corrected.

[jkrumbiegel]

@asinghvi17 wouldn't it maybe be a bit more intuitive if gutter gave the column width instead of half of it? Because mostly you're going to use this with categories anyway where 1 would mean touching and 0.8 a little gap, etc.

[asinghvi17]

@asinghvi17 wouldn't it maybe be a bit more intuitive if gutter gave the column width instead of half of it? Because mostly you're going to use this with categories anyway where 1 would mean touching and 0.8 a little gap, etc.

Hmm, I see what you mean. Since beeswarms are symmetric at this point I don't see why not, though that would call for a breaking release IMO.