Fix support for models that do support inbuilt function calling.

[palaniappan-r]

No description provided.

[Copilot]

Pull Request Overview

This PR removes support for inbuilt tool calling in models and refactors related code accordingly.

• Explicitly assigns a CPU device fallback in the FAISS vectorstore.
• Removes the inbuilt_tool_calling parameter from API router and agent instantiations.
• Renames and refactors the agent method to choose_tools in the RetrieverGraph.

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
backend/src/vectorstores/faiss.py	Explicitly assigns {"device": "cpu"} when use_cuda is False, ensuring a clear device configuration.
backend/src/prompts/prompt_templates.py	Removes the redundant tool_rephrase_prompt_template to simplify the prompt logic.
backend/src/api/routers/graphs.py	Removes the inbuilt_tool_calling parameter in line with the PR goal.
backend/src/agents/retriever_graph.py	Renames the agent method to choose_tools and removes the inbuilt tool calling branch.

Comments suppressed due to low confidence (2)

backend/src/vectorstores/faiss.py:34

• Consider adding a comment explaining why 'cpu' is explicitly assigned in the else clause to clarify the intent and improve code readability.

        model_kwargs = {"device": "cuda"} if use_cuda else {"device": "cpu"}

backend/src/agents/retriever_graph.py:123

• The renaming of the method from 'agent' to 'choose_tools' improves clarity; please ensure that any inline documentation or references to the old name are updated accordingly.

    def choose_tools(self, state: AgentState) -> dict[str, list[str]]:

[luarss]

LGTM