Henri, you added the dummy window tests, is there a chance you can explain what their purpose was? Do we care about the messages it receives?
IIRC the main point was to show/verify that creating a fullscreen Direct3D device makes the focus window the foreground window, even if the current foreground window belongs to a different thread. I.e., essentially commit 0838d0ab6491bdc0fb32d0b8715fbb57088b926a.
I don't think we care all that much about what messages the dummy window does or doesn't receive, although the "windowposchanged_received" bit introduced in 36553d862b539a95b71ec6fcea19f328e1562604 should probably take "hwnd" into account in any case. I think that's true for "syscommand_received" (and "wm_size_received" in the d3d9 version) as well.
Note that the window wasn't intended to be maximized either; that seems to have been introduced by 766e732fb37cf0c6dabd895f56506e5e4feb10cd. IIRC the idea there was to avoid resizing the device/focus window if it already matched the backbuffer size, but that doesn't really apply to the dummy window here. It should be fine to use e.g. a 100x100 dummy window as well.