Key realisations:
1. The driver entry point only needs to be called once per disk,
not once per bootable partition. Bootability is decided elsewhere.
2. As a result, my shim code does not need to access the disk at all.
3. None of the ROM-based ATADisk drivers support "slave" drives.
4. Apple drivers (contrary to the Monster Disk Driver Technote) only care
about D5 (the drive ID and some flags) on entry.
