Switch to "range" assignor strategy from "cooperative-sticky"

[daniil-quix]

Switched the partition assignment strategy to "range" from the previously used "cooperative-sticky".
With "cooperative-sticky", commits during rebalancing may fail, and we always commit the current checkpoint on assign and revoke.

According to this comment, range assignor now provides co-partitioning and ensures that the same partitions are assigned to the same members.

This property will help us to implement #484 too.

I briefly verified that it works this way, but I want to do more tests to be 100% sure.

Main changes

• Use "range" assignment strategy
• For recovery, subscribe to changelog topics too (previously subscribed only to input & repartition topics), and rely on assignment to be co-partitioned (same numbers to same consumers).
  • Added a check in case it's not co-partitioned
• RecoveryManager now doesn't assign/unassign partitions and instead only pauses, seeks and resumes them

[tim-quix]

Overall surprisingly simple/minimal changes, nice!

Besides my comments here, I had a very small PR #710 to remove some other redundant stuff that was prob just missed while you cleaned stuff up.