Optimize for reduction of LLM calls

[ghost-pep]

Currently, we make redundant calls to the LLM for things like claim extraction multiple times across different evaluators but for the same inputs. We should optimize this and potentially clean up this code.