Add teaching agent v2 training data and SageMaker fine-tuning notebook

Fix HuggingFace Spaces README metadata

Agent 2 implementation + baseline eval and small improvements

Agent 1 + eval