Improve kafka persistence

[terjokhin]

No description provided.

[Z1kkurat]

Didn't even notice this already existed, thanks @Daron666!
Short list of the changes in this PR:

• using new RebalanceConsumer1 from skafka
• some changes to core functionality by @nikitapecasa, I will take a look at them once more to ensure everything is in order and nothing more need to be done
• my rework of the current implementation of kafka-based snapshot persistence - the old one was reading the snapshots in the overriden PartitionFlowOf, now it reads snapshots with caching only once when KeysOf.all is invoked
• the large integration test mostly by @nikitapecasa ensuring that snapshot recovery works properly

[nikitapecasa]

Also this PR brings

  /** Safeguards a TopicFlow's resource to address a race between processing of messages and release of the resource.
    *
    * It provides following safeties in case of a fiber cancellation and resource's release:
    *
    *  - processing of current kafka messages won't be cancelled (TopicFlow.apply)
    *  - pending offsets would be committed
    *  - accumulated state would be persisted
    *  - resource would be released only after previous steps are completed
    *  - if resource is released before start of messages' processing, then it does nothing (no message processing, no commits, no persists)
    */

[nikitapecasa]

Also during working on this PR it was discovered that current graceful shutdown sequence can be improved #256