[Enhancement] Selective double-write in CombineConsumeQueueStore

### Before Creating the Enhancement Request

- [x] I have confirmed that this should be classified as an enhancement rather than a bug/feature.

### Summary

CombineConsumeQueueStore was originally designed as a transitional state in which every topic is double-written to both the file-based ConsumeQueue and the RocksDB ConsumeQueue. As it has gradually become a long-running state in practice, we observed that not all topics actually need double-write. In particular, LITE topics need it more than ordinary topics, because LMQ is required to run on top of the RocksDB ConsumeQueue.

This is the original motivation for introducing selective double-write.
To avoid an incompatible breaking change, we introduce a new opt-in switch `rocksdbCQSelectiveDoubleWriteEnable` as an additional control rather than altering the existing default behavior. Existing deployments are therefore unaffected unless they explicitly opt in.

### Motivation

.

### Describe the Solution You'd Like

.

### Describe Alternatives You've Considered

.

### Additional Context

.

Start by locating CombineConsumeQueueStore and the rocksdbCQSelectiveDoubleWriteEnable switch, then trace how topics are written to the file-based and RocksDB ConsumeQueues. Clarify the intended opt-in behavior for LITE and ordinary topics while preserving the existing default, and define tests that demonstrate both modes.

Written by the indexing model from the issue text.