[BUG] setting pixel tolerance too high can break multipolygons

[erik]

Describe the bug

Using a combination of Feature.setPixelTolerance(0.5) and Feature.setMinPixelSize(0.1) broke the geometry for this relation, which caused the entire tile to be filled at z11.

If you look at the relation, the sand and grass polygons are super close to each other (but not touching), so it seems the combination of calls resulted in an invalid simplification.

In my case, I tested that I can fix the tile by modifying the pixel size and tolerance values, but it seems possible for other multipolygons to have the same issue. The OpenMapTiles profile for example uses the same sequence of calls, and could conceivably run into this.

These changes all independently fixed the issue for me:

1. Remove the call to setMinPixelSize
2. Remove the call to setPixelTolerance
3. Change the pixel tolerance to 0.25

To Reproduce

• GeoFabrik's new-jersey extract.
• Create a layer which selects natural=sand polygons.
• Use the following profile:

    features
      .polygon("test")
      .setPixelTolerance(0.5)
      .setMinPixelSize(0.1)
      .setAttr("id", source.id())

• Relevant area: 11/40.9405/-74.0947

Commenting out either of setPixelTolerance or setMinPixelSize will fix the issue

Environment (please complete the following information):

• Hardware: 2019 Macbook Pro
• OS: MacOS 12.13.1
• Java version and distribution: openjdk-17
• Maven version: 3.8.2

[msbarry]

Interesting, looks like it's right on the edge of the buffer of tile at zoom 11, here's the geometry it generates:

POLYGON ((-4 -4, 100.375 -4, 100.5 -3.875, 100.625 -4, 260 -4, 260 260, -4 260, -4 -4))

So it might be an issue with clipping... I've got it reproduced in a unit test, should be able to investigate from there.

[msbarry]

Put out a proposed fix in #499 - let me know if there are other cases you can think of to test there?