Addressing custom box-box intersection

This does several things:

1. Modifies the "correctness" of box-box intersection:
  a. Addresses the interpretation of the contact position (as per
     issue #258); the contact position lies between the two surfaces.
  b. Correct the sign on the penetration depth so that colliding
     objects report positive penetration depth.
2. Refactors box-box intersection code:
  a. Removes redundant implementation to limit repeated bugs. This
     encompasses completely redundant function implementations as well
     as unnecessary duplicatio in branches.
  b. Add additional documentation/todos on the implementation.
  c. Replace the `fudge2` parameter with something more reasoned.
3. Adds unit tests confirming the behavior.