As per monarch-initiative/pheval.llm#49
Extraction responses (and others) may be fully empty - even beyond grounding or not, LLMs may return no content to be parsed.
These should not be cached.