[ODE] dt computation

[gl]

> This is trickier than it sounds.
> After some reflection, I've come to the conclusion that this is not
> desirable.
>
> For apps, and demos, using a proper dt is fine.
>
> However, the ode/test_ apps serve a different purpose: they should
> be testing ODE. And to test ODE, you will want deterministic
> behaviour. If each run of ODE is different due to timing, debugging
> becomes even more of a hell.

Was the suggestion to use a variable time step?  If so, why not implement 
the correct way of doing this, ie. accumulating elapsed time, and then using 
as many fixed steps as are required to catch up to it.  This is a FAQ anway 
and having it right in the demos would avoid future Qs.
--
gl