Add the Apple Silicon M4 bare-metal plan

The target is this machine: Mac16,10 / J773gAP / t8132. Records the full
SoC map read out of its own Apple Device Tree, and settles three claims
that the public record gets wrong.

SPTM does not block a third-party kernel. Apple ships two boot modes on
M4; the raw boot object mode still hands off at EL2 with the MMU off.
What SPTM broke is the reverse-engineering tooling, since XNU can no
longer be virtualised under m1n1. m1n1 already runs on an M4 Mac mini
and AsahiLinux/linux carries t8132-j773g.dts for this exact model.

The actual blockers are U-Boot, which has no t8132 case and so provides
no UEFI, and the Asahi installer, which cannot build a stub because no
IPSW old enough exists for this board. Both force direct m1n1 to kernel
entry over the Linux boot protocol, which is the kernel.bin lane the
Pi 500+ work already built.

Measured, and not published elsewhere: AIC v3 with its register layout
derived from the 0x4A00 die stride (max_irq = 4096) and an iack offset
of 0x40000 that independently matches Asahi's own t8132.dtsi; ethernet
as a BCM57762 that bge(4) already drives; and SPTM's presence in the
boot chain.

Corrects the folklore on three points: 16K pages are not required, as
OpenBSD and NetBSD both run 4K on Apple silicon; AICv3 already exists
under ISC in OpenBSD; and the console is DebugUSB/KIS rather than SBU
serial. Costs the critical path at ~3,100 lines, ~2,900 copyable under
ISC or MIT, with NVMe root avoiding the IOMMU because ANS sits behind
SART rather than DART.

Planning only. Nothing has been installed on the machine and no boot
policy has been changed.

Author: pkgdemon