Stop discarding in frag shader when coverage is zero and stencil writes are enabled

Stop discarding in frag shader when coverage is zero and stencil writes are enabled

This fixes the corruption in the NVPR images here:
http://108.170.217.252:10117/builders/Test-Win8-ShuttleA-GTX660-x86-Debug/builds/251/steps/CompareGMs/logs/stdio

caused by:

https://skia.googlesource.com/skia.git/+/65ee5f424cb4dabd453268902c00086605d77c1d

The stencil path step writes inside and outside the clip when the clip is implemented by
a coverage effect. The path cover step then doesn't write outside of the clip because the FS
discards. This leaves stencil values outside of the clip non-zero which messed up subsequent
path or clip draws to those samples.

R=robertphillips@google.com

Committed: http://code.google.com/p/skia/source/detail?r=13320

[bsalomon, 2014-02-04 22:05:33]

It's not clear to me why I thought this was necessary. It originates in this CL:
https://code.google.com/p/skia/source/detail?r=6741

I ran all the GMs in the msaa4, gpu, and nvprmsaa4 configs and nothing renders
differently (aside from fixing the NVPR corruptions). I modified GM to reuse the
same large target for all GMs to maximize the chance for cross-GM stencil buffer
corruption.

I'm going to submit some test try jobs just to make sure it doesn't break GMs on
other platforms somehow.

[robertphillips, 2014-02-04 22:33:33]

lgtm.

[bsalomon, 2014-02-05 15:58:38]

The CQ bit was checked by bsalomon@google.com

[commit-bot: I haz the power, 2014-02-05 15:58:47]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/bsalomon@google.com/154623002/30001

[bsalomon, 2014-02-05 15:59:20]

The N10 trybot GM failure is from the ToT code, not this patch.

[commit-bot: I haz the power, 2014-02-05 16:29:15]

Change committed as 13320