Compare fixed-size, semantic, and agentic chunking for RAG pipelines. Benchmarks, trade-offs, and when each strategy actually improves retrieval quality.