Side-by-side comparison of token-level LLM streaming using v1 graph.stream() and v2 StreamingHandler, both sync and async. Includes a TokenMetrics custom transformer to demonstrate extensibility vs the equivalent inline bookkeeping in v1.
