Make stencil and cover path rendering have more reasonable bounds.

src/gpu/batches/GrDrawPathBatch.cpp

Index: src/gpu/batches/GrDrawPathBatch.cpp
diff --git a/src/gpu/batches/GrDrawPathBatch.cpp b/src/gpu/batches/GrDrawPathBatch.cpp
index faf2d961fd220231958cc251d3b12ef440c24324..a829378e54f15d9ed82a8b5e0c5af2db8d5b3a82 100644
--- a/src/gpu/batches/GrDrawPathBatch.cpp
+++ b/src/gpu/batches/GrDrawPathBatch.cpp
@@ -56,18 +56,15 @@ bool GrDrawPathRangeBatch::isWinding() const {
 }
 
 GrDrawPathRangeBatch::GrDrawPathRangeBatch(const SkMatrix& viewMatrix, const SkMatrix& localMatrix,
-                                           GrColor color, GrPathRange* range, GrPathRangeDraw* draw)
+                                           GrColor color, GrPathRange* range, GrPathRangeDraw* draw,
+                                           const SkRect& bounds)
     : INHERITED(ClassID(), viewMatrix, color)
     , fPathRange(range)
     , fLocalMatrix(localMatrix) {
     SkDEBUGCODE(draw->fUsedInBatch = true;)
     fDraws.addToHead(SkRef(draw));
     fTotalPathCount = draw->count();
-    // Don't compute a bounding box. For dst copy texture, we'll opt instead for it to just copy
-    // the entire dst. Realistically this is a moot point, because any context that supports
-    // NV_path_rendering will also support NV_blend_equation_advanced.
-    // For clipping we'll just skip any optimizations based on the bounds.
-    fBounds.setLargest();
+    fBounds = bounds;
 }
 
 bool GrDrawPathRangeBatch::onCombineIfPossible(GrBatch* t, const GrCaps& caps) {