Bug: finish_reason='length' triggers spurious continuation retries on short complete responses (Ollama/GLM provider)

When the agent produces a short but complete final response, the finish_reason == 'length' truncation handler spuriously fires a continuation prompt, causing the same short response to be resent 2-3 times on Telegram.