kd: Use a more traditional k-d tree implementation

The slight extra pruning possible in the previous implementation didn't
seem to be worth it.  The new, simpler implementation is also about 30%
faster in most of the benchmarks.

This gets rid of Coordinate{Proximity,Metric} as they're not necessary
any more (and the old ExactNeighbors impl was too restrictive anyway).

1 files changed, 3 insertions, 3 deletions

diff --git a/src/lib.rs b/src/lib.rs
index d6e5579..57f3dac 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -464,10 +464,10 @@ pub mod tests {
 
     type Point = Euclidean<[f32; 3]>;
 
-    /// Test a [NearestNeighbors] implementation.
-    pub fn test_nearest_neighbors<T, F>(from_iter: F)
+    /// Test an [ExactNeighbors] implementation.
+    pub fn test_exact_neighbors<T, F>(from_iter: F)
     where
-        T: NearestNeighbors<Point>,
+        T: ExactNeighbors<Point>,
         F: Fn(Vec<Point>) -> T,
     {
         test_empty(&from_iter);