Makes Shell::SetDisplayWorkAreaInsets() call to WmShell

ash/aura/wm_shell_aura.cc

Index: ash/aura/wm_shell_aura.cc
diff --git a/ash/aura/wm_shell_aura.cc b/ash/aura/wm_shell_aura.cc
index 86158eb503c9122dd3da167994c98bb38eb2c25c..839854629fde8c032ed2f30aa9a67d1196725cc0 100644
--- a/ash/aura/wm_shell_aura.cc
+++ b/ash/aura/wm_shell_aura.cc
@@ -140,8 +140,9 @@ bool WmShellAura::IsForceMaximizeOnFirstRun() {
 
 void WmShellAura::SetDisplayWorkAreaInsets(WmWindow* window,
                                            const gfx::Insets& insets) {
-  aura::Window* aura_window = WmWindow::GetAuraWindow(window);
-  Shell::GetInstance()->SetDisplayWorkAreaInsets(aura_window, insets);
+  Shell::GetInstance()
+      ->window_tree_host_manager()
+      ->UpdateWorkAreaOfDisplayNearestWindow(window->aura_window(), insets);

[msw, 2017/02/16 18:34:49]

q: what's your advice on using WmWindow::aura_window() vs
WmWindow::GetAuraWindow()? Do you know that this will be non-null, or will we
just handle crashes as they come up? (I'm doing lots of WmWindow<->aura::Window
conversion in my WIP CL to eliminate WmWindowObserver, not sure I'll have a
reasonable CL to eliminate it in one fell swoop)

[sky, 2017/02/16 18:42:20]

Get rid of WmWindow! ;)

Good question. I think it depends upon the code. If the code may pass in null,
then you are right that WmWindow::GetAuraWindow() is safer.
The down stream code
(WindowTreeHostManager::UpdateworkAreaOfDisplayNearestWindow()) crashes on null,
so it seems fair to assume non-null here.

 }
 
 bool WmShellAura::IsPinned() {