Use the correct method for computing a repaint rect that takes into account squashing.

LayoutTests/compositing/squashing/tricky-element-removal-crash-expected.txt

 (GraphicsLayer
   (bounds 800.00 600.00)
   (children 1
     (GraphicsLayer
       (bounds 800.00 600.00)
       (contentsOpaque 1)
       (drawsContent 1)
       (children 1
         (GraphicsLayer
           (children 2
             (GraphicsLayer
               (position 8.00 8.00)
               (bounds 200.00 200.00)
               (contentsOpaque 1)
               (drawsContent 1)
               (backgroundColor #FA8072)
             )
             (GraphicsLayer
               (position 8.00 8.00)
               (bounds 100.00 100.00)
               (drawsContent 1)
               (repaint rects
+                (rect 8.00 8.00 92.00 92.00)

[chrishtr, 2014/05/30 16:58:14]

This one is I believe a chicken-egg issue  with layout and compositing ordering.
The actual repaint rect is (8 8 108 108), which is cut off here to the bounds of
the graphics layer. These bouds are recorded in the first layout pass of the
page, in which there are no graphics layers assigned yet, and hence the painting
backing of this div is the document.

I tested this hypothesis by modifying the test to teak layout in a no-op way
before removing the element. I verified that this causes the correct repaint
rect of (0 0 100 100).

                 (rect 0.00 0.00 100.00 100.00)
                 (rect 0.00 0.00 100.00 100.00)
               )
             )
           )
         )
       )
     )
   )
 )