[ODE] Stepfast bug fix

[Joe Ante]

> Hi!
> Is this patch reversed?  It looks like it probably is.  Thanks
> for sending a patch, but next time please send it as an attachment,
> since line-wrapping and tab-spacing have killed the patch's
> applicability when pasted into the body. :/
> 
> I would apply it by hand, but I'm not 100% sure about the
> (un)necessary zeroing of the 4th componant; I think we like to
> ensure sane real numbers in the 'padding' of vectors and matrices
> so that we can't accidentally keep doing maths on NaNs in the
> generic ODE matrix/vector ops and cripple our speed on archs
> which issue exceptions every time (PPC/OSX).  That's probably not
> an issue for various reasons, but I don't feel confident enough
> to apply that aspect of the patch without more opinions.
Would be nice to know whats the take on the redundant zeros from Russ.

As far as I can see it only affects anything when using altivec/mmx/sse.
Not sure about the pc side of things but altivec definitely doesn¹t throw
exceptions when using altivec to add nans or whatever.

Joe Ante