Process gzipped JSON incrementally
This saves heap space since we do not buffer the full content. Profiling org.wikimedia.discovery.cirrus.updater.common.graph.InflatingJsonResponseConsumerTest.profile
shows a difference of ~8GiB in allocation size when processing 10000 requests of 500KiB. CPU time difference is negligible, processing a response non-blockingly takes 1ms less than blockingly.
Bug: T355066 Change-Id: Ia5661a63056c7431e763100c8e16deb94ec0d176