RoundRobinLoadBalancer: increase visibility into health-checking

[idelpivnitskiy]

Motivation:

Currently, health-checking events are logged at DEBUG level and the
failure cause is not propagated. Users have to change the logging level
and reproduce in order to understand the root cause.

Modifications:

• Log each failed attempt to create a connection at INFO level for
  visibility;
• Log at WARN level when the health-checking is triggered;
• Retain the last cause that triggered health-checking inside
  HealthCheck state to use it for toString();

Result:

Users see when a connectivity issue occurs, cause is logged and
propagated for subsequent selection attempts if there are no other
ACTIVE hosts.