fix(clusters-dbscan): honor non-kilometer units in neighbor distance check

[monad98]

Issue

clustersDbscan ignores a non-kilometer units option. The bbox pre-filter honors options.units via lengthToDegrees(maxDistance, options.units), but the final neighbor predicate computes the distance in kilometers and compares it directly against maxDistance:

const distanceInKm = distance(point, neighborPoint, { units: "kilometers" });
return distanceInKm <= maxDistance;

So maxDistance is treated as kilometers in the predicate regardless of options.units. For any unit smaller than a kilometer this makes the check effectively always true, and the neighborhood is no longer clipped to a maxDistance-radius circle — it collapses to the square bounding box, clustering points that are farther apart than maxDistance.

Example: with { units: "meters" } and maxDistance: 400, two points ~566 m apart (400 m east + 400 m north) are clustered even though they are well beyond 400 m.

Fix

Measure the neighbor distance in options.units so it matches the unit used by both maxDistance and the bbox pre-filter. When units is omitted, both still default to kilometers, so existing behavior is unchanged.

Tests

Added a regression test covering a non-kilometer unit. All existing fixture tests still pass.

[mfedderly]

Oof, we definitely imply that it should work this way in the docs, but I'm afraid to merge this change without marking it as a breaking change. I've got a bunch of prep work for v8 staged, but haven't merged any of it quite yet. I added this to the v8 milestone though so that I should be able to include it there.

Thanks!