I've updated the MR with tests for scene state behavior after device reset for ddraw/d3d8/d3d9.
The only thing I'm unsure of here is where exactly to put the `wined3d_device_end_scene()` call in `wined3d_device_reset()`. Since it seems like it's effectively a noop right now the positioning in code currently doesn't matter, but if for some reason we decided to do something on calls to begin/end scene in the future this might be an issue.