We have golden tests for metal and vulkan, we should have them. As of 125180b8a7204de823544d658e89f7d3dac65e11 there were a handful of playgrounds tests that didn't render correctly and one that stalled the process.

I think the easiest way to implement this may be to use Angle on macOS to expand our existing golden test runner. If Angle works, I think the code will be almost exactly the same as we did for vulkan (see vulkan_screenshoter.mm).