No suspend/resume/shutdown lifecycle: controller loses programming across D3 while softc caches stay 'valid' — bus dead until reload

Summary

Neither attachment registers suspend/resume/shutdown: D3 resets the MMIO file but slave_valid=1/last_slave survive, set_slave_addr early-returns without reprogramming TAR/CTL/clock, and every post-resume transfer times out 25ms/byte until reload. All controller-behind devices (touchpads/sensors/SSIF BMC) silently die after suspend/resume. DEVMETHOD-resume re-running hw-init fix.