[ODE] why low mass geoms crash the simulation.

[Morgan McGuire]

> I'm always trying to understand why low mass geoms crash the simulation.

It may be because they are leading to non-positive definite inertia tensors.  I
found that in my (float32) system, dAddMass would create a non-PD result if the
mass was low.  I multiply small masses by 1000 before dAddMass and then divide
out later in order to correct for this floating point underflow problem.

-m