I think it would be compatible, in the same way a GPU would be compatible. You wouldn't be able to use it for your display though (for the same reason you cant use integrated graphics with your GPU).. so it would only be useful for accelerating ML or graphics rendering workloads for example.
Domain-specific accelerators have always had their own memory. Look at GPUs. I don't see why a power user who needs speed would be put off by some DMA.
Maybe Apple ends up implementing something like CXL down the line, anyway.