bug: AgentGraph hardcodes LLM and embedding models — ignores YAML configuration

[bhavyakeerthi3]

Bug Description

The AgentGraph class hardcodes the LLM to gpt-4o-mini and the embedding model to text-embedding-3-large. While the codebase has a YAML configuration system (config.yml), these parameters are not wired to it — making it impossible to switch models without modifying source code directly.

Impact

Any contributor or deployment that wants to switch models (e.g. Claude, Gemini, or a local Ollama model) must edit source code manually — making the existing YAML config system misleading since it implies configurability that doesn't exist for the most critical parameters.

Expected Behaviour

AgentGraph should read llm and embedding fields from config.yml and pass them through to get_llm() and get_embedding() — consistent with how other parameters are already handled by the config system.

Files Affected

• src/agent/graph.py
• src/util/config_yml/__init__.py
• config_default.yml

Note: Partially related to #112 — extends model configurability to parameters not covered in that PR.