Self-reflective Uncertainties: Do LLMs Know Their Internal Answer Distribution?