Replace cubic round rect corners with quads

Whenever Chromium needs an inverse RR clip it goes down this code path and gives us a path with cubics.

[robertphillips, 2013-10-16 14:49:57]

https://codereview.chromium.org/26372006/diff/2001/src/core/SkPath.cpp
File src/core/SkPath.cpp (right):

https://codereview.chromium.org/26372006/diff/2001/src/core/SkPath.cpp#newcod...
src/core/SkPath.cpp:901: 
Just moved up from below

https://codereview.chromium.org/26372006/diff/2001/src/core/SkPath.cpp#newcod...
src/core/SkPath.cpp:964: 
Not overjoyed with these parameters.

https://codereview.chromium.org/26372006/diff/2001/src/core/SkPath.cpp#newcod...
src/core/SkPath.cpp:1000: } else {
Want to avoid the extra "move" addArc would add.

https://codereview.chromium.org/26372006/diff/2001/src/core/SkPathRef.cpp
File src/core/SkPathRef.cpp (right):

https://codereview.chromium.org/26372006/diff/2001/src/core/SkPathRef.cpp#new...
src/core/SkPathRef.cpp:312: for (int i = 0; i < fPointCnt; ++i) {
This was firing due to floating point errors.

[f(malita), 2013-10-16 15:04:01]

As far as Blink side effects go, we should probably land this behind a flag -
then we're not tying up the Skia roll while handling a potentially major
rebaseline.

[robertphillips, 2013-10-16 15:17:35]

Now behind a flag.

[f(malita), 2013-10-16 18:01:04]

lgtm

[robertphillips, 2013-10-17 12:17:46]

committed as r11835

[robertphillips, 2013-10-17 16:29:57]

This was reverted in r11841 due to performance regressions

[robertphillips, 2013-10-21 12:57:37]

This was completed (regression fixed) in r11870. Please see:

Fix cubic->quad RR corner regression https://codereview.chromium.org/29673002/