[cairo] Seams with the image backend

[M Joonas Pihlaja]

On Sat, 2 May 2009, Soeren Sandmann wrote:

> Actually, looking more closely at the application, there is actually
> also a seam even when the slider is at 1.0 and the shared edge is
> specified with the exact same coordinates in the two polygons.

Yow.  I didn't expect that!

Turns out the problem is in the definition of the GRID_AREA_TO_ALPHA() 
macro:  It truncates rather than rounds when converting a span's area 
to [0,255].  I'll commit a fix to that soon.

After that fix if we decide that the image backend's seaming is still 
unacceptable in the collinear-but-not-exactly-equal-coordinates case, 
we might try doing the edge walking DDA in 24.8 space rather than the 
rasteriser's quantized space.  That might mean needing to increase the 
Y supersampling to 16x from 15x to keep the edge bucketing easy 
though.

Joonas