supervisord improvements

List of changes in this MR:

• better use of supervisord's API - we are no longer rewriting config file and letting supervisord reload it
• create Docker container
• improve performance (probably by implementing startup notifications)
• fix slow shutdown (takes exactly 10 seconds every time)

---

Performance

Using manager with supervisord is way slower than with systemd. Startup with systemd takes around 0.7sec on my system, while with supervisord it's nearly 4 seconds. This is caused by supervisord's method of detecting successfully started services. A process is considered started, when it is still running startsecs seconds after startup. There is an open issue about this for 7 years. We need some kind of successful start detection, otherwise our safety measures against invalid config do not work, so we can't turn it off. Systemd has systemd-notify, which gives us reliable way to detect successful startup faster.

This is a big problem on multi-core systems. With 32 kresd's, a change of configuration would take more than 32 seconds. I consider this unacceptable. The solutions to this are:

1. find some workaround
2. patch supervisord
3. implement systemd-notify or another way of startup notifications in manager
4. find a replacement for supervisord that handles this better as there is no technical reason why we couldn't use anything else
5. decide, that we can ignore this

I think that patching supervisord or implementing some form of state tracking ourselves will be the way forward. Or just letting it be and fixing it later.