SATA hotplug connect tasks are unsynchronized and never drained at channel teardown: racing ATTACH/DETACH tasks yield NULL-softc panic or duplicate disk children

Summary

Every PHY-changed interrupt allocates an independent task on the per-CPU taskqueue - nothing serializes two tasks (concurrent on different CPUs) and no nata detach path drains pending ones: teardown frees+NULLs the softc so the task body NULL-derefs, and racing ATTACH/ATTACH both run ata_add_child creating duplicate disk children. Hardware-triggered (PHY flap or admin kldunload racing hotplug) - no unpriv trigger. Softc-NULL guard + teardown-drain fix.