[cairo] Analytical curve bounds

[M Joonas Pihlaja]

On Mon, 29 Dec 2008, Behdad Esfahbod wrote:

> Thanks Kalle.  Since all the results are in thee 20% difference range I wonder
> how much of it is noise.  In fact, I wonder how much of the tests actually hit
> the new code path at all.

All the fills with curves do hit the new code path when rendering 
with the image backend.  I was thinking we might want to go back 
to control point extents there since they don't _need_ to be very 
accurate, and the change to use the approximate_fill_extents was 
by accident.

Joonas