Fine-Grained Detection of Context-Grounded Hallucinations Using LLMs