feat: introduce base_url_fixed for local model

[TaeBbong]

Summary

This PR adds support for configuring base_url separately for llm_engine and llm_engine_fixed, enabling users to use local LLM servers (e.g., vLLM) for all components.

Problem

Previously, the base_url parameter was only passed to llm_engine, while llm_engine_fixed always used the default API endpoint. This caused APIConnectionError when:

1. Running with a local vLLM server without external API keys
2. Using model_engine = ["trainable", "trainable", "trainable", "trainable"] configuration

The error occurred because:

• Planner.llm_engine_fixed and Verifier.llm_engine_fixed were created without base_url
• Base_Generator_Tool also lacked base_url parameter support

Solution

Added new parameters to construct_solver():

• base_url_fixed: For llm_engine_fixed (falls back to base_url if not set)

Modified Files

• solver.py: Added base_url_fixed and base_url_tool parameters
• planner.py: Added base_url_fixed parameter for llm_engine_fixed
• verifier.py: Added base_url_fixed parameter for llm_engine_fixed
• initializer.py: Pass base_url to tool instances
• tools/base_generator/tool.py: Added base_url parameter support

Usage

VLLM_BASE_URL = "http://localhost:8000/v1"

solver = construct_solver(
    llm_engine_name="vllm-Qwen/Qwen2.5-7B-Instruct",
    model_engine=["trainable", "trainable", "trainable", "trainable"],
    enabled_tools=["Base_Generator_Tool", "Python_Coder_Tool"],
    tool_engine=["self", "Default"],
    base_url=VLLM_BASE_URL,           # For llm_engine
    base_url_fixed=VLLM_BASE_URL,     # For llm_engine_fixed
)