Fix: blank/truncated chat replies from thinking models
Thinking models served via ollama's OpenAI-compatible API (e.g. gemma4:12b) routinely return an empty message.content with the actual answer in a message.reasoning field. ctk's provider read only content, so chat replies came back blank or cut off (in the TUI, a query could produce no visible output at all).
Fix: capture reasoning onto ChatResponse.reasoning and fall back to it when content is empty, in both chat() and stream_chat(). Replies are no longer silently blank, and the reasoning is now available structurally for richer rendering later.
Note: this surfaces the model's raw chain-of-thought as the reply text when content is empty. Polished rendering (thinking shown distinctly from the final answer, a live progress indicator, and cancel) is in progress.
pip install -U conversation-tk