I think the problem here is not necessarily the AllowNVIDIAGPUScreens option itself, but the fact that it requires a ServerLayout section to specify it. There’s a known issue with the X server where it will prefer a server layout it finds in xorg.conf.d/* over one you specify yourself in /etc/X11/xorg.conf. So even though your layout named Layout0 has two X screens, the server is using the one named layout that it’s finding in /usr/share/X11/xorg.conf.d/10-amdgpu-nvidia-drm-outputclass.conf that has an implicit one-screen layout. This causes the server to create a GPU screen for the other Device section you specified in xorg.conf.
I think you’ll need to delete the ServerLayout sections in /usr/share/X11/xorg.conf.d/* before your custom layout will be used. Alternatively, you should be able to start the X server with the -layout Layout0 option to choose your layout explicitly.
A future driver release will enable AllowNVIDIAGPUScreens by default on xserver 1.20.7 and newer so hopefully this problem will go away relatively soon.