High zoom tiles are too dark

[e-n-f]

High zoom tiles (with buckets of around z22 or higher) are too dark because a lot of points are snapped to roads instead of reported at their natural locations.

Possible fixes:

• Add positional noise during encoding
• Detect and discard exact duplicate locations
• Make better use of density quantiles to highlight lower densities at these zooms
• Just change the curve at these zooms instead of bothering with the quantile calculation

[e-n-f]

This is even more of an issue when absolute maximum densities for each zoom instead of kll.

Positional noise doesn't work: it makes everything fuzzy and causes irregular jumps in brightness between zooms because of the artificiality.

Using density quantiles won't work because using kll to calculate them runs out of memory when tiling the world to high zooms.

Maybe the thing to do is to just use the z0 max brightness, which is going to be pretty honest for any dataset that covers an area larger than a z9 tile, and hardcode the per-zoom ramp from there. Possibly the gamma should also go up as the zoom levels do, to keep sparser areas of the world visible as you zoom in.

[e-n-f]

It is easy to calculate the maximum enclosing tile for the data, so it will work better for data confined to smaller geographic areas to normalize against the maximum density in that zoom rather than z0. (Still will be bad for, say, a London map, which isn't contained within any tile other than z0.)