[To dev/1.3] Harden RegionMaintainer retries and region operation atomicity

[CRZbulabula]

Scope

This is the partial-fix delivery for TDB-280 / PingCode V1-465 (V1.3.8.1). It does not claim full resolution.

Changes

• Ported the positive RegionMaintainer, DataNode RPC, and idempotent state-handling increments from the final master state without reintroducing RemoveRegionGroupProcedure.
• RegionMaintainer groups tasks by full region ID, preserves per-region FIFO, and blocks later tasks behind a failed head task.
• The persistent queue polls only SUCCESS, CREATE + REGION_ALREADY_EXISTS, and DELETE + REGION_NOT_EXIST. Failed tasks remain for the next retry round.
• Added asynchronous DELETE_REGION handling and explicit DataNode idempotent statuses; real errors remain failures.
• Combined consensus-peer and local-region deletion results idempotently; DeleteDatabase enqueues region deletion before partition metadata cleanup and routes synchronous deletion failures to the async queue.

Tests

• ConfigNode focused tests: 26 passed.
• DataNode focused tests: 14 passed.

Limitations

Full procedure state persistence, checkpointing, recovery, and compensation semantics are not included in this partial fix.