[Fix] Raise ValueError when llm-eval gets an unknown dataset to evaluate on (instead of failing silently)

[Fix] Raise ValueError when llm-eval gets an unknown dataset to evaluate on (instead of failing silently) #5354