Same approach as in GeoGraphicsScene. Placemarks are assigned a tile
level that fully contains them once. When clipping to some boundary
box, only those placemarks on the path towards the top of the pyramid
can intersect them. Note that this is only a heuristic; checking
the actual bounding box intersection again afterwards still rules
out a lot of placemarks that don't actually intersect.

Rough speed comparison for different approaches:
A) Slow:      clip all placemarks directly
B) Medium:    clip all placemarks based on tile pyramid
C) Fast:      clip all placemarks that intersect with the bounding box
D) Very fast: clip all placemarks based on tile pyramid that also
              intersect with the bounding box

C) is the old implementation, B) something I tested locally and C) the
new implementation.