Some minor tweaks to accumulateNodeDirtyRegion for performance.

 - After calling computeDirtyRegion, the dirtyRegionTemp is updated with the same minX, maxX, minY, maxY of the returned BaseBase. However in many cases (I think every case other than a perspective transform) the dirtyRegionTemp is the same instance as the returned bb, and thus we can skip this assignment entirely
 - Possibly even better than doing this check in accumulateNodeDirtyRegion would be to change the implementation of computeDirtyRegion so that it flattens in-place and doesn't return a BaseBounds object at all. Note sure if this helps.