[REVIEW] fix rag following latest cline upgrade

[hugolatendresse]

• Now able to get Aria to try to invoke search_actuarial_tools
• Updated to Gemini's gemini-embedding-001 (their only active embedding model). . Anthropic doesn't not expose embedding models.
• We are not constrained by only being able to use a Gemini key for everything. As long as a cline account is connected and the user also has provided a Gemini API key to cline, we can use Claude models for the chatbot and the vector search will use Gemini.
• Also deleted the old RAG logic we don't need anymore.

Don't forget:
npm cache clean --force
npm run install:all
npm audit fix
npm run protos

---

Rebuilt RAG index with:
source venv/bin/activate
cd /home/hugo/code/aria
pip install -r requirements.txt
cd rag
python export_index.py --rebuild

---

Below is an example of RAG answer obtained from the Werner textbook while using claude-sonnet.