Create floating-point square root path for 16b mirrored radial gradients

Create floating-point square root path for 16b mirrored radial gradients
(6x-16x speedup).  Create specialized dy=0 path for clamped radial gradients
(25% speedup).  Make 16b radial gradient proc signature similar to 32b,
passing in SkScalar instead of SkFixed and/or SkFloat.

git-svn-id: http://skia.googlecode.com/svn/trunk@3030 2bbb7eff-a529-9590-31e7-b0007b416f81