Hybrid GPU-CPU CAGRA Indexing

Milvus 2.6.1 introduces a hybrid GPU-CPU approach for CAGRA indexes, leveraging GPUs for accelerated graph construction and CPUs for cost-efficient, scalable query execution. This addresses the trade-offs between GPU-native CAGRA's build performance and its query-serving cost and scalability constraints. The `adapt_for_cpu` parameter allows flexible configuration for different operational scenarios, enabling scenarios like GPU-built CAGRA serialized and queried as HNSW on CPUs for cost-sensitive workloads.