[server] Fix thread-safety bug in RebalanceManager ZooKeeper recovery path

[fhan688]

Purpose

RebalanceManager.initialize() is invoked on the startup thread but currently calls registerRebalance() directly, which mutates non-thread-safe state owned by the coordinator event thread. This violates the single-threaded execution contract and opens a race window during coordinator HA failover or restart.

Brief change log

• Add RecoverRebalanceEvent, a new coordinator event that carries the RebalanceTask read from ZooKeeper during startup recovery.
• Change RebalanceManager.initialize() to enqueue a RecoverRebalanceEvent via eventManager.put(...) instead of directly calling registerRebalance().
• Handle RecoverRebalanceEvent in CoordinatorEventProcessor to delegate registerRebalance() onto the coordinator event thread.
• Add testStartupQueuesRecoverRebalanceEvent, which verifies that startup() enqueues the recovery event rather than executing the rebalance inline.

Tests

• org.apache.fluss.server.coordinator.rebalance.RebalanceManagerTest#testStartupQueuesRecoverRebalanceEvent — verifies the recovery event is enqueued instead of registerRebalance being called inline.
• Existing RebalanceManagerTest suite (5 tests) — all pass, no regressions.

API and Format

No.

Documentation

No.