Trying to force ever more restrictive and intrusive controls upon players won't solve cheating. The only way to "solve" cheating is with
https://xkcd.com/810/. Use statistical analysis and server-side controls (fog of war, lockstep calculations) to force cheaters to play indistinguishable from top human players. If you can't tell the difference, does it even matter?
> the types of dongles used for this are likely detectable in some way by kernel-level anticheat, and computer vision based cheats probably work better when you can inject contrasting color textures into the game
If you've ever worked in broadcast or volunteered for conference, lecture or house of worship broadcasting, you'll know there's an entire industry of cheap undetectable HDCP-removing HDMI splitters and capture cards. It's an open secret that conference AV relies on shitty $10 chinese HDMI splitters to make HDCP "work".
Similarly, there's a countless number of devices that can present themselves as any other USB device. You can MitM e.g. a keyboard or controller and inject packets that are impossible to distinguish from the users' own inputs.
Some consoles only allow wireless controllers with encrypted protocols, but that can be circumvented too. Replacing the joysticks in controllers with hall-effect ones is a common mod. It's possible to attach another chip inbetween at this point to inject custom inputs.
You can use these injected inputs to e.g. compensate for recoil. But you can also run a simple classifier on the HDMI video to identify objects and players.
Now sure, an anti-cheat could use statistical analysis to measure how quickly a player reacts, which would allow detecting such cheats. At this point it won't matter whether you're using kernel, userland or server-side anticheat though, as they've all got the same information available to them.