Address some bubble clipping issues

ui/views/window/non_client_view.h

Index: ui/views/window/non_client_view.h
diff --git a/ui/views/window/non_client_view.h b/ui/views/window/non_client_view.h
index 1eac8c29ead65def7368481a52652dbc11da35ce..899d0292aaf924b4163216a5ec07ec7d58676f33 100644
--- a/ui/views/window/non_client_view.h
+++ b/ui/views/window/non_client_view.h
@@ -65,6 +65,9 @@ class VIEWS_EXPORT NonClientFrameView : public View,
   virtual gfx::Rect GetWindowBoundsForClientBounds(
       const gfx::Rect& client_bounds) const = 0;
 
+  virtual bool GetClientMask(const gfx::Size& size,

[tapted, 2016/03/22 07:38:04]

I think this is kindof a nice abstraction (it should have a comment :), but
could BubbleFrameView::Layout() just call
GetWidget()->client_view()->set_clip_path(...); ? Maybe that's simpler, but
maybe it's also nice to have this as a method on NonClientFrameView, so it's
clear any NonClientFrameView can clip its ClientView - up to you (or sky/msw :)

[Evan Stade, 2016/03/22 17:59:58]

That approach is fine with me, but it requires BubbleFrameView::Layout() to have
a side effect outside of its own child hierarchy, which seems like some sort of
violation. I'll let sky/msw comment on which approach they prefer. Agreed that
this function needs a comment.

+                             gfx::Path* mask) const;
+
   // This function must ask the ClientView to do a hittest.  We don't do this in
   // the parent NonClientView because that makes it more difficult to calculate
   // hittests for regions that are partially obscured by the ClientView, e.g.