Fix socket specs when network not available

[straight-shoota]

Socket specs were failing when running on a system with no network.

UDP multicast specs are already pretty fragile and handle many points of failure as acceptable because the host system/network might not support UDP multicast. This patch just adds code for another failure type.

Then there are a couple of specs which initiate a DNS lookup to a non-existent domain (doesnotexist.example.org.). When no DNS server is available, getaddrinfo apparently returns EAI_AGAIN to try again later (which of course won't help anything when there's simply no network to reach an upstream DNS server).
I'm not entirely sure if simply accepting EAI_AGAIN is the best way to fix this. The spec could also be marked as pending in that case. But I think success is fine.
Also, I'm wondering if EAI_AGAIN should be handled somewhere in Addrinfo, retrying and finally raise a more specific error.
But that's a whole different story. For now I think these changes should be good to make the spec suite work on isolated machines (a use case for that is the build environment on build.opensuse.org/).

If you want to run specs without network access, I found the easiest way is a docker container with --network none.

[beta-ziliani]

The Windows job is failing

[beta-ziliani]

I think marking them as pending instead of success makes more sense.

[straight-shoota]

Yes, that's happening in UDPSocket spec.

The other specs are already expecting failure, they're just picky about the error code. I think accepting EAI_AGAIN as a "successful" failure makes sense. There's no way to tell exactly what caused it, it just indicates "temporary failure". And that should be good for these specs.