Add a socket timeout option to handle lingering socket requests on transitioning or mixed state nodes [JIRA: CLIENTS-624]

[mk1s]

There have been several recent problems with nodes in various states transition whether it be physical or service(Riak) being up where the riak-ruby-client will try and connect for upwards of 60s or more, until if fails and tries the next node in the node list. Some sites have been able to fix this by tweaking DNS. Suggested recommendation was to add an option to define a socket level timeout to provide more predictable behavior.

[seancribbs]

Thank you for the bug report @mk1s.

Implementation detail:

1. Add SO_TIMEOUT option on the PB socket, expose appropriately.
2. Check whether Excon can support SO_TIMEOUT or other timeout-y options.
3. Add support for timeout.rb-style SystemExit exceptions in Client#recover_from. (Customer request)

[boosty]

We added PB timeout support to our custom fork of riak-ruby-client:
https://github.com/xing/riak-ruby-client/commit/67f28e1f66e22f79191668aa7d7978d6007f7d89

We will send pull requests soon, but need some time to rebase our fork against the current master.

[sbimikesmullin]

+1 to get this merged upstream

[boosty]

I created a PR for setting custom socket timeouts: #265

[lukebakken]

Will be resolved by #270