Fixes #1172 - Add per-runId message buffering in ChatLog - Separate thinking stream from content stream handling - Ensure proper sequencing (thinking always before content) - Model-agnostic: works with or without thinking tokens