Actually make sleepBeforeConnect() implement an exponential backoff

[ewust]

Currently, our sleepBeforeConnect() doesn't sleep for the first 6 connections, but then sleeps 1 second between retries after that.

That's too overwhelming, and not an exponential backoff. This change increases the exponential backoff to be 25*(2^n) milliseconds, up to 15 seconds, for retry n.

We may want to even increase the upper bound (15 seconds) to something larger (or remove the upper bound entirely)

[jmwample]

I am happy to merge this as it is. I think we should consider adding config values to replace the constants (25 for the ramp, and 15s for max in this PR). These could even potentially go into the clientconf as a deployment setting so that clients have a more easily tunable backoff if things start to get overwhelmed. We might open an issue for that relating to #100