task_wdt: HW fallback is not triggered when corrective actions fail

[fschuette-gcx]

Consider a scenario with the task watchdog having configured multiple channels which are fed quite regularly and the hardware fallback being activated. One channel stops feeding, the watchdog is triggered and its corrective actions (e.g. reboot) are about to be executed. Assuming that, for some (unlikely) reason, the corrective actions fail, the hardware fallback watchdog will not be triggered, as other channels continue to feed it. IMO this behavior should be modified so that the hardware watchdog is not fed once the task watchdog initiates corrective actions.