Dataset format refactoring

Refactor dataset for RAFT training.

Before	After
question: `str` User requested question.	id: `str` ID of self (QA Dataset item)
prompt: `str` Prompt actually delivered to model.	dataset_id: `str` ID of Dataset
doc_ids: `list[str]` IDs of documents retrieved from vector store.	question: `str` User requested question.
positive_doc_ids: `list[str]` IDs of documents which distance is below threshold.	context: `str` Retrieved answers from vector store.
negative_doc_ids: `list[str]` IDs of documents which distance is above threshold.	reason: `str` Chain of Thought response explaining answer from model.
reasons: `str` Dumped python `dict` containing metadata of `doc_ids`.	answer: `str` Answer response from model.
answer: `str` Answer response from model.
output: `str` Full output of reasons and answer.

junnoslab / DQChat-LangGraph