Fix empty response from reasoning/thinking models

[tkdlabs]

Problem

When using a reasoning ("thinking") model — e.g. many models served via Ollama — the agent message box renders completely empty and no response is shown.

Cause

Such models stream their reasoning as delta.reasoning_content, during which delta.content is None. The streaming loop in chat.py treats any non-string content chunk as the end of the stream and breaks immediately on the very first chunk, so the entire response is discarded before the model's actual answer arrives:

if isinstance(chunk_content, str):
    ... append ...
else:
    break   # fires on the first reasoning chunk

Fix

Skip non-string content chunks with continue instead of break. The async iterator still terminates naturally at end-of-stream, and the model's answer (which follows the reasoning) renders as normal. One-line change + explanatory comment.

Verified against several thinking models on a local Ollama server (gemma/qwen/glm families): empty boxes before, correctly streamed answers after.