I was thinking a bit more about what might be required. If a center camera is required, there isn't an extra camera input in the V3 AP processor. So Tesla could need a new 3.1 AP processor with a ninth camera input. It would also need new wiring from the headliner to the AP processor for the new camera. This solution seems unlikely.
Another approach would be to replace the windshield triple camera with the new 2 front camera HW4 design and then have the cabin camera as well. That solves the wiring issue (3 cameras still going to the AP processor. This has its own set of issues, as I suspect Tesla only needs two front cameras with the new high-resolution cameras. Of course that may require HW4 AP processor to process the 4X+ data from each camera! Still, it might be possible for the V3 AP processor to handle the higher-resolution cameras in a clever way.
Anyway, I can see this being a can of worms for the developers - trying to work out a solution that can be economically implemented with the least amount of new hardware, yet take advantage of the massive amount of video data used to train the neural networks.