Don't draw objects when not in the view frustrum. Somewhat related to #81.

Notes:

• Note that shadows can still be cast inside the frustrum.
• Multiple frustrums can be combined and vectorized using the new linalg.
  • When rendering, the rendering camera frustrum and all shadow camera frustrums should be evaluated together (first test which shadow camera frustrums intersect the rendering frustrum) against all bounding volumes in the scene, so that the set of objects can be computed that affects the visual end result.
• There will be objects that don't need culling (e.g. backgrounds), and objects for which we cannot know the bounding volume; these should be excluded from the test and always be selected for rendering.