Reliable test job

[dkfellows]

One of our tests was disabled because it was not generating reliable results (and had become steadily more broken over time; code that isn't tested at all is code that can be going horribly wrong). This is about trying to make it run and be more reliable.

[dkfellows]

Reliability increased by:

1. ensuring that worker threads in the mock server have actually started
2. defining and using a connection object (wrapping the connected socket) so that a server can have multiple connections open at once
3. making the match for the number of keepalives less precise; the exact number is not important, so long as they're sent.

[dkfellows]

I've not yet removed this comment:

	/*
	 * TODO test need fixing as calls to client are time dependent so this test
	 * sometimes fails. For example j.getState() may or may not call the client
	 * depending on when the previous state was requested.
	 */

I'm not sure if it applies or not.

The sensitivity was due to things like the caching period being unconfigurable. Fixed that (just for testing).

[dkfellows]

That failure up above was a transient in networking unrelated to our tests.

[dkfellows]

If this causes failures, I'll be very sorry and fix it then.