[Youwes09]

DAPF init was skipped on T8132 because it raised an SError. dapf_entries
hardcodes each DART's DAPF reg index, but on T8132 dart-pmp, dart-dcp and
dart-dcpext0 have no DAPF, and the reg at that index is a PS_PROT page shared
by many DARTs.

Changes

• pmgr: pmgr_adt_power_is_active() reports whether a device and the
  parents it depends on are powered. Virtual gates (dart-ane's on T8132, under
  an ANE_SYS that is off at boot) have no state of their own.
• Find each DAPF from the ADT. Walk every DART under /arm-io and take the
  DAPF from its instance property, in either layout: {type, unit, name[8]}
  per reg range (the range of type DAPF), or {type, name[8]} per DART range
  (an entry of type DAPF puts the DAPF in the range after them). The entry
  count tells them apart. The table is now only a fallback for DARTs with no
  instance. No chip checks.
• Skip a DAPF with no filter data. On T8132, T8142 and T8152, dart-ave has
  a DAPF but no filter data; it was reported as an error.
• Power the device behind the DART. The gates are on the DART or on the
  device named without dart- (isp, ane, ave, pmp). Power it only if it is
  off and restore it after; if its state cannot be read, power it and leave it
  on rather than write unpowered.
• Drop the T8132 skip in kboot_boot().

ADT collection

Simulated against every ADT in GravityLinux/adt-collection; each DART that
gets initialized is exactly one whose instance lists a DAPF:

SoC	DAPFs initialized
T8103	aop, ave, isp, ane
T8132 (J713, J604)	aop, mtp, isp, ane (ave: no filter data)
T8132 (J773)	aop, ane (ave: no filter data)
T8140	pmp, mtp, ave, isp, ane
T8142	mtp, isp, ane (ave: no filter data)
T8152	ane, ane1 (ave: no filter data)
T6040	aop, ave0, ane0
T6041	isp0, mtp, aop, ave0, ave1, ane0
T6050	pmp0, pmp1 (no filter data), ave0, ave1, ane0

The table missed ave0/ave1/ane0 on T6040/T6041, pmp0/pmp1 on T6050 and ane1
on T8152.

Testing

J713 (Mac16,12, 13" M4 Air), macOS 26.6.2 firmware, booting Linux 7.3: no
SError, keyboard and trackpad work.

A debug build read each DAPF's registers before and after init:

• Before init, every DAPF holds random contents with the enable bit clear:
  iBoot leaves no configuration to preserve.
• After init, every filter reads back as written, except that the end address
  is rounded up to a 4-byte boundary (...000 reads back ...003).
• The ANE DAPF keeps its filters across a full ANE_SYS power cycle, so
  initializing once in m1n1 is enough.

Not tested on other SoCs.