Add a hack-around to this fatal taskqueue running whilst the NIC

is detaching.

This mostly fixes a panic - the reset path shouldn't run whilst
the NIC is being torn down.

It's not locked, so it's "mostly" ok, but most of the rest of
the driver doesn't read sc->invalid with sensible locking. Grr.

The real solution is to cleanly tear down taskqueues in the detach/suspend
phase, but ..

sys/dev/ath/if_ath.c

@@ -2318,6 +2318,9 @@ ath_fatal_proc(void *arg, int pending)
 	u_int32_t len;
 	void *sp;
 
+	if (sc->sc_invalid)
+		return;
+
 	device_printf(sc->sc_dev, "hardware error; resetting\n");
 	/*
 	 * Fatal errors are unrecoverable.  Typically these