Dispatcher threads should not be blocked forever

[mdbergmann]

Dispatcher threads (aka actors with a threaded message-box) currently can be blocked forever, or for a long time, when certain tasks are submitted.
That shouldn't happen.
It should be possible for the dispatcher to monitor the threads and destroy threads that are hung in order to spawn new threads, based on a configuration option.

[mdbergmann]

Each dispatcher should have a 'sentinel', a separate thread that monitors the dispatcher workers.
A worker records the time when it starts working so that the sentinel can check how long that task is going on.
Based on a configurable threshold the sentinel can either report work that exceeds the threshold, or also kill the thread.

[mdbergmann]

It should be up to the user to not submit long running work, but the framework should provide means to at least report suspicious stuff.