Scale-in nodes failover compact the StreamSetObjects

[superhx]

Who is this for and what problem do they have today?

Why is solving this problem impactful?

Compact the StreamSetObjects of the scale-in nodes with splitThreshold=1MiB :

• Most of the data: Most of the data is split into StreamObjects to achieve independent TTL and better data cohesiveness.
• Remaining 5000 partitions * 1MB: The remaining data will be periodically scanned and eventually expire through StreamControlManager#cleanupScaleInNodes.

Additional notes

[superhx]

We need to prepare more CPU / Memory resources for compaction failover.

https://github.com/AutoMQ/automq-for-rocketmq/tree/kafka_issues706

Implement it when you have a more elegant solution.