socket: support network IO deadline (#339)

[grrtrr]

This resolves #339 by setting an r/w timeout for pipe reads/writes
(if present). This allows listening sockets to recover from bad
incoming connections, and also goroutines suffering from broken
connections, to recover.

[grrtrr]

TestPubSubIPC seems different on Windows, on my Debian 9 system it runs:

=== RUN   TestPubSubIPC
--- PASS: TestPubSubIPC (1.07s)
    common_test.go:589: Address ipc://mangostest20477, 6 Cases
    common_test.go:158: Jan 30 12:32:24.772: Id 0: Got all 0 messages
    common_test.go:158: Jan 30 12:32:24.779: Id 4: Sent all 0 messages
    common_test.go:158: Jan 30 12:32:24.779: Id 3: Sent all 0 messages
    common_test.go:158: Jan 30 12:32:24.779: Id 2: Sent all 0 messages
    common_test.go:158: Jan 30 12:32:24.779: Id 1: Sent all 0 messages
    common_test.go:158: Jan 30 12:32:24.779: Id 5: Sent all 0 messages
    common_test.go:158: Jan 30 12:32:24.829: Id 0: Sent all 7 messages
    common_test.go:158: Jan 30 12:32:24.829: Id 0: Testing complete
    common_test.go:158: Jan 30 12:32:24.829: Id 4: Got all 2 messages
    common_test.go:158: Jan 30 12:32:24.829: Id 3: Got all 2 messages
    common_test.go:158: Jan 30 12:32:24.829: Id 5: Got all 2 messages
    common_test.go:158: Jan 30 12:32:24.829: Id 1: Got all 2 messages
    common_test.go:158: Jan 30 12:32:24.829: Id 1: Testing complete
    common_test.go:158: Jan 30 12:32:24.829: Id 2: Got all 2 messages
    common_test.go:158: Jan 30 12:32:24.829: Id 2: Testing complete
    common_test.go:158: Jan 30 12:32:24.829: Id 3: Testing complete
    common_test.go:158: Jan 30 12:32:24.829: Id 4: Testing complete
    common_test.go:158: Jan 30 12:32:24.829: Id 5: Testing complete

[gdamore]

I want to take a completely different approach here. The use of a deadline to workaround the fact that we are doing blocking operations during an accept call isn't really a fix but a bandaid. What we need to do is have the negotiation steps take place on a separate go routine before we add the pipe to the socket.

Unfortunately, this is going to require some rather significant surgery to the plumbing. I hope to get to this soon, but I'm honestly buried in some other stuff in NNG (and $dayjob) at the moment, so it may be a week or more before I get to this.

[grrtrr]

I appreciate your commitment to ongoing improvement of this software.
The pull request tried to go into a different direction, I will close this for now and sum up the thoughts in the issue; if your time permits within the next weeks, I'd be grateful for feedback.