[ArtemTetenkin]

Fix Kafka node registration race on Orion startup

Problem

When Orion starts, the cluster is registered with ClusterManager right after its sensors are started, but before the first KafkaBrokerSensor run (describeCluster) has returned. During that window, agent heartbeats hitting /register can create broker nodes via Cluster.updateNodeFromAgentHeartbeat, using the agent-reported NodeInfo as the node's currentNodeInfo.

Fix

• KafkaCluster overrides updateNodeFromAgentHeartbeat and defers registration of new nodes until the first broker sensor run has populated the node map. Heartbeats for nodes that are already known (added by the sensor or restored from cluster state) are processed as before.
• KafkaBrokerSensor calls markBrokerSensorCompleted() right after brokers are added to the node map, so a later failure (e.g. controller lookup) doesn't keep registrations blocked.
• Deferred heartbeats return null (no command), which agents already handle; they retry on their next heartbeat. Metrics from the heartbeat are still published.
• To avoid blocking registrations forever if Kafka is unreachable at startup, the gate is lifted after a 2-minute grace period.