Restart single boxes by debug box

To avoid denial of service of a complete application by attacking one single box it is necessary to restart single boxes by the debug box.

For further explanation of the issue/feature here some quotes of the mbed forum:
[Reinhard Schweiger:](https://forums.mbed.com/users/Iceberg)
> I have a question about the reaction of uVisor after detecting a tried access to a restricted area or a missing ACL. Right now, during my first tests, I only saw a complete stopped OS on my target after an unallowed access. The reason for my question is that I want to build applications which at an unallowed access of a box only stopp this box while all other boxes keep running.
> This reaction would be nessesary in an application which handles a sensor and an actuator in one box and a website in another box for visualizing the value of the sensor. If somebody injects code to the box of the website to overwrite the values of the sensor, the box with the website should be stopped and the box with the sensor and the actuator should keep running.
Is it possible to stop only the box with the unallowed access? Where can I configure this reaction?

[Milosch Meriac:](https://forums.mbed.com/users/meriac)
> Current uVisor reboots the device after catching a security problem. The solution for just rebooting a single box would be implemented using the planned [debug box](https://github.com/ARMmbed/uvisor/issues/70) feature. The debug box would be able to catch a box faulting and resetting just that box.
Architecturally its better to move the logic for handling a box-specific reset into the application debug box, as it can notify depending boxes of the reset, as they would need to re-establish potential session.

This was discussed here: [ Change uVisor reaction at hitting missing ACL and unallowed access?](https://forums.mbed.com/t/change-uvisor-reaction-at-hitting-missing-acl-and-unallowed-access/1920)

Start with the planned debug box feature in issue #70 and the linked forum discussion. Determine how a box fault should be caught, reset, and reported to dependent boxes without rebooting the device. Done means the faulting box is restarted while other boxes continue running and can re-establish any required sessions.

Written by the indexing model from the issue text.