[ODE] performance with rays

[Erwin de Vries]

> oops. growing the AABB wouldn't work. i think to avoid extra checks, the
ray needs to be broken up into pieces.  perhaps the step could be specified
when you create the ray.  the step would be 0-1 as a fraction of the total
length of the ray.
>
> in addition, a ray 'radius' would be useful and easy to implement because
at the point of potential contact, you just check a sphere radius instead of
an epsilon.
>
>   dGeomID dCreateRay (dSpaceID space, dReal length, dReal radius = 0.f,
dReal step = 1.f)
>
> comments?

Wouldnt it be wiser to just implement the CCylinder class then? This is a
fat ray.

Erwin