fix(chatbot): add test coverage for all 17 intents; fix dataset quality

Wave 2: compare_destinations data + NER fixes + evaluation update

eval(chatbot): Wave 1 retrain + held-out evaluation results