test: do not use valid public hosts for timeout testing

[Trott]

A couple of assumptions I've made:

• Purpose of the test/internet directory: Is it just tests that rely on an internet connection but really shouldn't ideally? Or is that an incorrect assumption? (Is it documented anywhere?) Now that this one doesn't rely on a working internet connection, should it be moved to test/parallel instead?
• DNS lookup vs. no DNS lookup seems irrelevant in the context of testing socket timeout so I removed the DNS lookup one. (I couldn't figure out a reasonable way to do it that didn't involve internet connectivity. It needs to do a DNS lookup, after all.) Am I missing something or does that seem about right?

Additional information that may be useful:

• 240.0.0.0 might, theoretically, be used at some future time. But that is unlikely. And even so, it's still a better choice for the foreseeable future than a valid Time-Warner IP, which is what it seems the previous one was.
• There is a "discard" prefix address in IPv6 that would be useful here, but I couldn't get it to work. I'm reluctant to file a bug report because the problem is likely just my IPv6 ignorance.

[Trott]

CI: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/80/

[brendanashworth]

Is it just tests that rely on an internet connection but really shouldn't ideally?

I think they are tests that can rely on one. If it doesn't rely on one anymore, I figure it could be moved to parallel.

[bnoordhuis]

You could replace the callback with just assert.fail.

[bnoordhuis]

LGTM

[Trott]

landed in c786d63

[Fishrock123]

Shouldn't this test be moved to /parallel/ then?

[brendanashworth]

@Fishrock123 see #2257 😄