Ternary VSA Reasoning: A Learned Multi-Hop Reasoning System with Vector Symbolic Architectures and Ternary Weights

Version 2 - fully rewritten with: per-hop accuracy breakdown (1-hop 100%, 2-hop 100%, 3-hop 70%), system architecture documentation (PredicateShardedStore, FastController, GraphKANResonator, ChainScorer, NL Parser v2), component memory footprints, hybrid regex+neural NL parser with 100% entity accuracy, and MCU export details for Cortex-M0+.