[FEA] Construct HNSW hierarchy from a CAGRA graph on the CPU

[cjnolet]

This is being done currently inside of FAISS to enable interoperability between CAGRA (GPU) and HNSW (CPU). The greedy search algorithm from HNSW is used to take the CAGRA base graph, which is flat, and construct an HNSW hierarchy from it so that vectors can continue to be added to the index. We need to do this also for HNSWlib inside of cuVS, as not all users of CAGRA are using FAISS. Initially, doing this on the CPU should be okay, but we need to work towards doing this on the GPU in the longer term. I'm going to create a separate Github issue for that.