KafkaIndexTask maxRowsInMemory is per-Sink, should be not-per-Sink

[gianm]

Since there is up to one Sink per time range per partition, you can actually potentially get a lot more rows in memory than you expect. Would be nice for it to apply across all Sinks.

[navis]

seemed related to #2459, but that is for realtime task

[gianm]

should be fixed by #3284 for appenderator/kafka.

@navis thanks for pointing out #2459, will take a look.

[fjy]

👍 we should fix this