Dicklesworthstone/frankensearch

# Technical Summary Provides both a library and standalone CLI (`fsfs`) for local-only hybrid search, with a modular crate architecture separating embedding, lexical indexing (Tantivy), vector storage (FSVI with quantized f16), and fusion orchestration—enabling independent evolution of each layer. Queries flow through canonicalization and type classification, then execute a two-phase pipeline: fast BM25 + potion-128M embedding fused via Reciprocal Rank Fusion for immediate results, followed by optional quality reranking with MiniLM and cross-encoder integration. Environment variables, config files, and CLI flags control phase behavior (fast-only mode, RRF constants, scoring weights), with streaming output modes (JSON Lines, TOON) designed for agent integration and watch-mode incremental indexing for corpus updates.