RoundRobinLoadBalancer: further logging enhancements

[idelpivnitskiy]

Motivation:

There are still a few gaps in RRLB logging.

Modifications:

• Add a counter for targetResource to distinguish different clients
  for the same target host;
• Enhance RoundRobinLoadBalancer#toString() to include all hosts;
• Use consistent "Load balancer for" pattern for all log events;
• Avoid double logging when health check passes (it also logs that it
  cancelled);
• Log the full host state when health check is cancelled to let us
  know its new state;
• Log connection id if we close a new connection when health check
  finishes;
• Log if the health check terminated with an unexpected error;
• Fix order of log arguments when we observe connect failure;

Result:

Better visibility into RRLB execution.