Use a ConnectionPool instead of ConnectionWrapper for the SystemManager

[wildstray]

Useful for working on clusters with nodes down (eg. with Cassandra Cluster Admin)

[thobbs]

I'm hesitant to do this right now, because it can easily result in schema disagreements if your nodes don't have their clocks synchronized with NTP and you're running something lower than Cassandra 1.1.

However, once it has been a while and most people are using 1.1 or newer, this would be nice to have. For now, I suggest closing the SystemManager object upon failure and opening a new one (possibly against a different node).