enhance: support different replica numbers on secondary CDC cluster

[bigsheeper]

Summary

In a CDC (Change Data Capture) replication setup, the primary and secondary Milvus clusters may have different hardware resources (different number of QueryNodes, different resource group topology). Currently, when a LoadCollection is replicated from the primary to the secondary via CDC, the secondary blindly applies the same replica_number and resource_groups from the primary. This can cause load failures (not enough nodes) or resource waste (over-provisioning) on the secondary.

Proposal

Allow the secondary cluster to use its own cluster-level replica configuration (ClusterLevelLoadReplicaNumber / ClusterLevelLoadResourceGroups) instead of the primary's replica config when processing replicated AlterLoadConfigMessage.

Approach

1. Proto change: Add use_local_replica_config bool field to AlterLoadConfigMessageHeader
2. ReplicateService: Set the flag to true on every replicated AlterLoadConfigMessage on the secondary
3. DDL callback: In LoadCollectionJob.Execute(), when the flag is true, read local cluster-level config and substitute it for the primary's replica config. Fall back to primary's config if local config is not set.

Key Design Decisions

• No new config parameters - reuses existing ClusterLevelLoadReplicaNumber / ClusterLevelLoadResourceGroups
• Backward compatible - when local config is not set (default), falls back to primary's config
• No changes to primary cluster behavior - only the secondary side is affected
• Covers TransferReplica too - also uses AlterLoadConfigMessage, preventing primary from rearranging secondary's resource groups

Data Flow

Primary: LoadCollection(replica_number=3, rgs=["rg1","rg2","rg3"])
  → WAL: AlterLoadConfigMessage(replicas=[{rg1},{rg2},{rg3}])
  → CDC replicates to secondary

Secondary ReplicateService:
  → Sets header.use_local_replica_config = true

Secondary DDL callback (LoadCollectionJob.Execute()):
  → Sees use_local_replica_config == true
  → Reads local ClusterLevelLoadReplicaNumber = 1
  → Reads local ClusterLevelLoadResourceGroups = ["__default_resource_group"]
  → Rebuilds replicas = [{__default_resource_group}]
  → Spawns 1 replica in default RG

Affected Files

File	Change
pkg/proto/messages.proto	Add use_local_replica_config field
internal/distributed/streaming/replicate_service.go	Set flag on replicated AlterLoadConfig
internal/querycoordv2/job/job_load.go	Read local config when flag is set
Tests	Unit tests for both components