In-process approximate-nearest-neighbour CacheStore for Kmemo, the semantic cache for LLM calls on Kotlin/JVM — a pure-Kotlin HNSW index that scales the default store past an exact scan, with no extra dependency.