[Pixman] Radial gradients (again)

[Andrea Canciani]

I don't remember the timeline for pixman 0.20, but I think that the
attached patch should be reviewed and merged (if it looks good),
because it is a straightforward change that extends the valid domain.

http://cgit.freedesktop.org/~ranma42/pixman/commit/?h=wip/radial&id=ce1738313e0a3e98bc004f1b27e5644929f2427c

I've been working on a fragment-shader implementation of the
new gradients. This might be a good starting point for fast paths
using SIMD instructions, because there is no branching.

http://cgit.freedesktop.org/~ranma42/cairo/log/?h=wip/gl2

In the same branch I also tried to extend one of the tests to
hit all the important cases (in particular tangent circles and
constant radius were not tested at all).
I want to port this test or something very similar to pixman
and check it against a different (and slow, but numerically
stable) implementation of the gradients, but it will take some
time, because I will be quite busy until the end of October.

Andrea