Kinematics should not be added to constraint batch handle sets

[RossNordby]

Given that a kinematic body's velocities cannot be modified by constraints, there is no danger in allowing multiple references to the same kinematic body within a constraint batch.

This can be important- imagine five hundred dynamics sitting inside a big kinematic spaceship. Without making this change, there would be five hundred constraint batches (at least until #15 exists, but even if it did, there's no good reason to trigger that fallback with kinematics).

This is pretty easy to implement- the only difficulty is keeping everything synchronized when a body shifts from kinematic to dynamic. All of its constraints would need to be reallocated.

[RossNordby]

With sleeping/waking now implemented, the complexity penalty for handling this special case has increased a bit.

Given that the jacobi fallback solver is essentially required anyway, I'm leaning towards just relying on it for this. Only consider doing this after the jacobi fallback is implemented, and only if performance demands it.

[RossNordby]

With the jacobi fallback implemented, I'm going to drop this for the foreseeable future; it's a lot of complexity for vanishingly little benefit.