pub fn thought_message(reasoning: &str) -> MessageExpand description
Build a model-role Message carrying model reasoning as a
ThoughtContent, NOT as answer text.
The reasoning rides one ThoughtSummaryContent text part. Renders
downstream as a collapsed “thinking” line (TUI LineKind::Thought) and is
kept out of the assistant’s reply. Used for providers that stream reasoning
separately (e.g. z.ai GLM’s reasoning_content). The control plane prunes
reasoning from the replayed prompt (it is never replayed to the provider).