Connection pooling

[cavaliercoder]

Connection pooling is currently recommended via PG Bouncer/other (http://cavaliercoder.com/libzbxpgsql/documentation/connection-pooling/).

There is an opportunity however to reuse a single connection for each agent thread. This would mean that:

• connection overhead is significantly reduced
• server version information can be persisted between each query to simplify version switching of queries

We would need:

• a hash of the connection so multiple connections can be monitored per thread
• a way to query the state of a connection and reconnect if needed
• to pass server information to each metric handler so it knows which version to query
• reasonable timeouts
• cleanup on exit