The call to IDirect3DDevice9::Present in testDeviceLost sometimes never returns on AMD, resulting in a hang.

Also, on AMD, device lost is sometimes never reported at all, even through present occluded or display mode changed. I'll need to do something more robust at the app level. It has something to do with the Chrome GPU process sandbox. These issues only happen when the sandbox is enabled.

Cleanup for Surface::recreateAdditionalSwapChains to follow once I determine this gets us back to a stable place.
Review URL: https://codereview.appspot.com/8038044

git-svn-id: https://angleproject.googlecode.com/svn/trunk@2040 736b8ea6-26fd-11df-bfd4-992fa37f6226