Implement two step shutdown with sigint

[ocaballeror]

Fixes #60

Here's a quick rundown of the changes:

• The signal handler needs to send a shutdown signal to the agents from a separate thread, since running it in the same one as the nameserver's socket (default), would lead to a deadlock where the nameserver is waiting for the agents to shut down and the agents are waiting for the name server to accept their 'unregister' message.
• The nameserver process can't call terminate on itself from the sigint handler because ._popen seems to be undefined when accessed from a different process. Right now it seems the NameserverProcess is being terminated anyway (the tests would block otherwise, and there's no zombie python process afterwards), but I'm not entirely sure how or why.
• I removed the sigint handler on the AgentProcess. I don't think it's needed anymore, since we are capturing the signal from the nameserver.
• This took me way longer than I expected, so I hope this is the right way to do it 👍

[codecov]

Codecov Report

Merging #249 into master will decrease coverage by 0.2%.
The diff coverage is 94.73%.

@@            Coverage Diff             @@
##           master     #249      +/-   ##
==========================================
- Coverage   99.15%   98.94%   -0.21%     
==========================================
  Files          26       26              
  Lines        3550     3619      +69     
  Branches      256      259       +3     
==========================================
+ Hits         3520     3581      +61     
- Misses         18       24       +6     
- Partials       12       14       +2

Impacted Files	Coverage Δ
osbrain/agent.py	98.38% <100%> (+0.67%)	⬆️
osbrain/tests/common.py	100% <100%> (ø)	⬆️
osbrain/tests/test_agent.py	100% <100%> (ø)	⬆️
osbrain/nameserver.py	97.84% <86.66%> (-1.38%)	⬇️
osbrain/tests/test_nameserver.py	97.6% <94.73%> (-0.52%)	⬇️
osbrain/tests/test_bugs.py	83.33% <0%> (-16.67%)	⬇️
osbrain/__init__.py	91.66% <0%> (-8.34%)	⬇️
osbrain/tests/test_agent_transport.py	93.97% <0%> (-2.41%)	⬇️
osbrain/proxy.py	98.75% <0%> (-1.25%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ad5e4e2...c03435f. Read the comment docs.

[ocaballeror]

More changes:

• Re-added and fixed the sigint handler for agents.
• Added a test for the agent's sigint handler based on the commits from this branch: https://github.com/Flood1993/osbrain/tree/sigint_tests
• Added psutil as a testing dependency. It makes process management much easier to handle, especially if you are dealing with multiple OSs.

@Peque Let me know how much you disagree with me 😄

[ocaballeror]

@Peque Is this a good moment to go on with this?

This should be rebased on top of the appveyor integration, by the way, I don't trust Windows signals whatsoever.

[Peque]

@ocaballeror Let us first improve the current Windows CI integration (combine coverage reports, fix tests currently failing, see if we can skip less tests...).

Trying to tackle this now would only add more noise to the currently incomplete Windows support.

[ocaballeror]

@Peque Shall we take a look at this again?

[Peque]

@ocaballeror Sure. I will review this this week.

[Peque]

@ocaballeror Before reviewing this, would you mind rebasing against master?

[ocaballeror]

I didn't realize Windows compatibility was still not a thing when we started implemented this, so it will require a bit of work.

What I found from a few minutes of playing around with this is that:

• Windows doesn't interpret a ^C event as a signal.SIGINT, but rather as a signal.CTRL_C_EVENT, which is kind of unfortunate.
• Windows doesn't support os.wait() either, so we will probably have to work around this by using psutil.Process.wait().
• Something weird happens on test_nameserver_sigint_kill(), since it's reporting a PermissionError on os.kill().