Cache JMX connection and MBeans names and attributes

[max-melentyev]

Hey, I saw significant improvement with jmx connection cached. Here is a chart for sample cassandra cluster without any traffic, orange line is for an instance with patched exporter:

Cassandra Clusters with traffic have GC time and throughput decreased, and we saw latencies improved up to 50%. But this depends on the traffic patters and vary from cluster to cluster.

Implementation details:

• scraper is cached in collector's config instance, so it's replaced every time config is reloaded.
• jmx connection is cached in scraper, and it uses Cleaner API to close it when scraper is discarded.
• MBeans names and attributes are cached in scraper. JMX client subscribes to MBean register/unregister events to invalidate this cache.
• JMX is reconnected when any exception occurs. I tested only in javaagent mode, so didn't have a chance to test reconnection.

[dhoard]

@max-melentyev Thanks for the PR! I'll need to review it.