Summary

• detect suspicious Ollama-hosted GLM finish_reason="stop" responses after tool turns when the visible text does not end naturally
• reclassify that narrow case to "length" so the existing continuation path can request the missing tail instead of silently ending the turn
• add regression coverage for the misreported-stop path plus scope guards for complete responses and non-Ollama backends

Testing

• python3 -m pytest -o addopts= tests/run_agent/test_run_agent.py -k "ollama_glm_stop or non_ollama_stop or length_finish_reason_requests_continuation"

Closes #10711