ata-isa registers no device_detach and never unwinds probe-allocated resources: kldunload of the ISA nata attachment permanently fails (ENXIO) and attach-failure leaks both IOPORT ranges with dangling softc pointers

Summary

The methods table omits detach (the shared ata_detach exists and every sibling attachment wires it) - devclass_delete_driver aborts at the kobj-error ENXIO while any ISA channel is attached, and failed attaches leak both IOPORT ranges for the boot with stale r_io vectors. Root/boot-hint gated; GENERIC compiles nata in so the unload consequence additionally needs a module build. Wire-detach + release fix.