Safir_control must not be multithreaded due to fork() issue

[DonOregano]

We have seen a few crashes where safir_control and/or dose_main crash with a message about "epoll reregistration" on Linux. It seems that on POSIX OS:es it is very difficult to be thread safe and do fork()/exec() calls, which is what safir_control currently does.

The asio documentation mentions the threading issue: http://www.boost.org/doc/libs/1_61_0/doc/html/boost_asio/reference/io_service/notify_fork.html

Also there are a few relevant stack overflow discussions:

https://stackoverflow.com/questions/21529540/how-do-i-handle-fork-correctly-with-boostasio-in-a-multithreaded-program

https://stackoverflow.com/questions/9362258/how-to-make-boost-asio-fork-safe

Simplest resolution is to make safir_control single-threaded.

[DonOregano]

Fixed, by making it single-threaded.
An issue might be that Communication and/or System Picture may start misbehaving.
In that case the solution might be to use the trick in one of the SO discussions above: fork a child while still single threaded and let it do the fork.
This latter solution might be desirable when we handle all processes.

[DonOregano]

This discussion appears very relevant:
klemens-morgenstern/boost-process#65