Have Appveyor run nanomsg_sys tests; Fix nanomsg_sys tests on Windows.

[GGist]

In reference to #132 and #133.

I realized that Appveyor was not running tests for nanomsg_sys so I have fixed that. The results from that commit can be seen here. You can see that tests::constants_should_match_return_of_symbol_func() failed and tests::should_create_a_pubsub() ended up blocking.

I fixed tests::should_create_a_pubsub() by increasing the sleep time and fixed tests::constants_should_match_return_of_symbol_func() by changing more Windows specific constants. The passing build can be found here.

@blabaere In regards to the comment left on my previous commit, my intuition when it comes to the differences between C runtimes is lacking. If I understand correctly, the constants can vary between operating systems and compilers? I know that in the two main compilers for Windows (gcc via mingw and msvc) both of the errno.h constants were identical. I can't speak for how they might vary in other circumstances.

[blabaere]

From what I can remember, the problem occurred when nanomsg was compiled with a version of the C runtime, and then an application using nanomsg was compiled with another version having different values for the constant. I'll try to find the original explanation of the problem. It has to be somewhere on the nanomsg site, or in the code comments or in Martin Sústrik blog's.

Now that we have a CI on windows and a test covering that point, I'm quite confident that the code is now correct, even if I'm not sure where the problem was coming from.

Thanks for the fix !