Characterising LLM-Generated Competency Questions: a Cross-Domain Empirical Study using Open and Closed Models