feat: Assistive chatbot v2: decompose questions and evaluate Guru card retrieval

[yoomlam]

Ticket

Resolves https://navalabs.atlassian.net/browse/DST-184: Build pipeline from question-transformer LLM to retriever

Changes

Build pipeline from question-transformer LLM to retriever, including retrieval evaluation

• original question → transformed question(s) → retriever → retrieval results; compare against desired Guru cards
• In a sense, the LLM is a tool-using or function-calling LLM, where the tool is the retriever, and the tool input is a question.

Testing

Testing instructions and expected behavior:

1. pip install -r requirements.txt
2. Update .env depending on the LLMs you want to test
3. Run LLM_MODEL_NAME="gpt-4-turbo" RETRIEVE_K=4 ./decompose-questions.py

[yoomlam]

Moved these into functions to reduce load time. Lazy-load them (i.e., only when needed).