[Bugfix] Preserve retained subscription progress after the last consumer closes

[Caideyipi]

Closing the last consumer currently deletes its group's checkpoints even when the topic has retain.progress.after.unsubscribe=true. Recreating the group can then start at the WAL tail and skip data written while the group was offline. A delayed asynchronous queue close can also recreate detached WAL retention after retention was disabled, the topic was dropped, or a replacement queue was registered.

This change preserves configured checkpoints and the empty group's incarnation and SQL dialect on ConfigNode and DataNode, so resubscription can resume correctly. Empty table groups encode their dialect using a negative empty-consumer count, preserving the existing serialized field boundaries. Disabling retention or dropping the topic still clears retained progress and prunes empty ConfigNode groups once no checkpoints remain.

WAL retention registration now tracks the owning queue. Delayed closes unregister the old queue but cannot restore a canceled or superseded detached registration.

Validation:

• 88 targeted unit tests passed, covering group serialization, selective checkpoint cleanup, snapshots/rejoin in both data models, and deterministic asynchronous-close races.
• Full English and Chinese reactor test-compile -DskipTests -Dxml-format.skip=true passed (Chinese with -P with-zh-locale).
• The new IoTDBConsensusSubscriptionBasicTableIT#testLastConsumerCloseRetainsProgress passed on 1 ConfigNode + 1 DataNode with IoTConsensus, covering both direct close and unsubscribe-then-close, followed by consumption of offline writes without seek. Subscription was enabled only in temporary test artifacts because the community edition disables it at compile time.
• Spotless and git diff --check passed.