Major Description
Since we check if the partition is fetchable or not in fetchedRecords(), there is a bug when the partition is paused during there is in-flight fetch request, it will not be ignored in fetch response handler but after that, in fetchedRecords(), causing the fetcher to update the fetched position already; later no fetch requests will ever be sent to the broker for this partition since consumed != fetched.
The proposed fix is to move this check from fetchedRecords to handleFetchResponse in Fetcher.