feat: Add LLM evaluators (#3571)

* Spike out LLMConcisenesseEvaluator (and more generic CriteriaEvaluator)

* Add test using llm evaluator

* Add EvaluatorFactory and define initial set of LLM-based evaluators

* Tweak type definitions

* Clean up factory-generated classes

* Improve class construction

* Update notebook to use LLM evaluator

* Use more robust label -> score mapping logic

* Address PR comments

* Use new model name

* Use inline comment for longevity

* Use different evaluator to avoid unique constraint on name in tests