Support for KIP-219: Throttling

[vctoriawu]

While librdkafka has support for a throttle_cb set by the application code, would the best practice be to throttle all requests in the application layer or in librdkafka?

If all requests are throttled in the application layer, the application will need to know what are requests that need to be throttled vs not, and this can change if additional APIs are added. However, this would prevent the throttle_cb from having to re-enter into librdkafka code.

If requests are throttled in librdkafka, there would need to be a function call from the throttle_cb in the application layer back to librdkafka. All requests could be throttled from a single place in the code. Is this something worth exploring and contributing to?

Thanks!

[vctoriawu]

@edenhill any thoughts?