Throttle collision computation to save performance on low-end devices

[wipfli]

On low-end devices, it might be beneficial to reduce the rate at which we run collision computations. I think at the moment, MapLibre recomputes label collisions during zooming. To save some CPU cycles, we could recompute the collisions only once a zoom event has completed.

So for example rater than doing 24 times “zoom 0.1”, “recompute”, “zoom 0.1”, “recompute”, …, we could do “zoom 2.4”, “recompute”.

If we only recompute collisions after the zoom event, we need to scale the text label size together with the linear scale of the map, otherwise you get visible collisions…

I think the behavior described here is roughly what google maps does in some versions that are optimized for low-end devices. At least this is what google maps does in the iOverlander app.

Would be great if we could test something like this and have some performance metrics that we could use for evaluation.