Sometimes, when connectivity issues happen, SDKs produce an enormous amount of logs describing transient issues.

E.g. this happened over 48 days when the service experienced connectivity issues, resulting in ~500 INFO log records from AmqpChannelProcessor alone.

While we can (and should) adjust logging levels in #20836, it should also be useful to throttle logs at specific threshold configurable by users (can be off by default or on at some enormous numbers). It would provide a safe belt in case of unexpected issues and won't make them worse because of extra logging.
We can also write a warn log when throttling starts every once in a while.

The concern here is that throttling can be a logging implementation feature and we need to investigate if sinks provide existing solutions.