rapidsai/cuvs-lucene

Implements a custom `KnnVectorFormat` codec that delegates vector indexing and search to NVIDIA's cuVS library, offloading computation to GPUs for accelerated approximate nearest neighbor operations. Built as a pluggable Lucene codec requiring CUDA 12.0+ and JDK 22, it integrates directly into Lucene's codec pipeline without modifying core search logic. Enables GPU-accelerated vector operations within standard Lucene indexes while maintaining compatibility with Lucene's codec architecture.