cardturner

refactor: simplify model response header logic to ensure clean separation from thinking blocks

feat: add streaming LLM responses with optional thinking output and configurable tracing verbosity

feat: Begin turning cards