feat: implement DiskANN index with PQ support

[huanglune]

Description

Implement DiskANN (Disk-based Approximate Nearest Neighbor) index to support large-scale vector search that exceeds memory capacity.

Features

• DiskANN index implementation with disk-based graph storage
• Product Quantization (PQ) for vector compression
• Aligned I/O infrastructure for efficient disk access
• KMeans clustering for PQ codebook training

Technical Details

DiskANN Index

• Disk page layout for vectors and graph edges
• Builder for constructing disk-based index
• Searcher for efficient disk-based ANN search

Product Quantization

• PQ codebook training using KMeans
• PQ encoding/decoding for vector compression
• Distance computation using lookup tables

Infrastructure

• Aligned reader for O_DIRECT disk I/O
• Memory alignment utilities

Related

• PR: feat: add DiskANN index with PQ support #65