Cull features from places layer based on grid

[nvkelso]

• What did you see? There are a ton of locality points in the places layer starting at zoom 8 when they come in from OSM until around zoom 13. We source those from OSM and they have kind_detail values (eg city, town, village) which we map to min_zoom, and most have population values. But there are sometimes hundreds when just a handful can end up being labeled. We can't select by population ranked via zoom because that would leave the more rural areas of the map looking empty. So instead we need to cull features based on a grid that roughly matches how many features we could actually label in a given tile.
• What did you expect to see? A smaller number of features based on visual layout of what can be labeled. At zoom 8 this might be ~12 features total (for visual layout reasons not just tile size reasons), then zoom 9 to 24 features, zoom 10+ to 36 features? And by zoom 13 no more culling? Maybe this is accomplished with different grid cell sizes at different zooms, too?
• What map location are you having problems with? Tile 8/227/100 in 256px coordinates
• Screenshot? This tile from OSM.org shows Tokyo, which has tons of locality points at that zoom.
  

That same area via Overpass Turbo:

Helpful queries:

/*
This has been generated by the overpass-turbo wizard.
The original search was:
“place=town and population=*”
*/
[out:json][timeout:250];
// gather results
(
  // query part for: “place=town and population=*”
  node["place"="city"]({{bbox}});
  way["place"="city"]({{bbox}});
  relation["place"="city"]({{bbox}});
  node["place"="town"]({{bbox}});
  way["place"="town"]({{bbox}});
  relation["place"="town"]({{bbox}});
);
// print results
out body;
>;
out skel qt;



/*
This has been generated by the overpass-turbo wizard.
The original search was:
“place=town and population=*”
[out:json][timeout:250];
// gather results
(
  // query part for: “place=town and population=*”
  node["place"="city"]["population"](if:t["population"]<5000)({{bbox}});
  way["place"="city"]["population"](if:t["population"]<5000)({{bbox}});
  relation["place"="city"]["population"](if:t["population"]<5000)({{bbox}});
  node["place"="town"]["population"](if:t["population"]<5000)({{bbox}});
  way["place"="town"]["population"](if:t["population"]<5000)({{bbox}});
  relation["place"="town"]["population"](if:t["population"]<5000)({{bbox}});
);
// print results
out body;
>;
out skel qt;
*/

[nvkelso]

This can be implemented as a post processing step, similar to what we do here to "keep_n_features" station from the pois layer:

• vector-datasource/queries.yaml

  Lines 1415 to 1422 in 15a8fb3

  	# only keep all stations below z12, then cut it down to the top few.
  	- fn: vectordatasource.transform.keep_n_features
  	params:
  	source_layer: transit
  	start_zoom: 9
  	end_zoom: 12
  	items_matching: { kind: station }
  	max_items: 20

That function is implemented here:

• https://github.com/tilezen/vector-datasource/blob/2eb4486761062f67d473424e6526d1ebf7ba7643/vectordatasource/transform.py#L3117-L318

I recommend duplicating that function and adding grid logic to it, with the grid spacing configurable. If it goes well we could later migrate the existing "keep_n_features" transfoms in queries.yaml to the new functions.

[nvkelso]

This zoom 7/113/50 parent tile of the example above at 512 pixel size has 724 places features, of which all but 8 are locality points.

50.mvt.zip

It should only have a few dozen features total.

[nvkelso]

GeoJSON from Overpass for approximately the same area to use for tests (most of those properties don't need to be kept):

• japan_places_layer_locality_features.geojson.zip

[nvkelso]

Closed via #2001.

[nvkelso]

Followup in #2040.