[Enhancement] Support tools in LLM Node

[psriramsnc]

🚀 Support Tool Calls in LLM Node via New ModelResponse

🧩 Summary

This PR introduces comprehensive support for tool calls in the LLM Node by introducing a new ModelResponse abstraction. It enables better agent-style evaluation, structured responses, and multi-modal compatibility — all while maintaining backward compatibility and clean integration across components.

---

⚙️ Features Implemented

• 🆕 Added ModelResponse class — All custom models now return a standardized ModelResponse object from the generate_response method.
• 🧠 Tool call support in LLM Node — Enables evaluation of agentic model behaviors with native tool execution flow.
• 🪄 Graph post-processor support — Integrates graph-based processing for flexible response handling.
• 🌐 Enhanced HTTP Client — Supports json_payload as a boolean flag and allows custom headers via model config.
• 🖼️ Multi-modal test message support — Ping messages now supports multimodal formats by default.
• 💬 Fixed multi-turn chat history injection — Properly injects user, assistant, and tool roles in conversation state.
• 🧰 Added convert_openai_to_langchain_toolcall utility — Simplifies tool call conversion for chat history injection.

---

🚄 Performance Impact

Tool calls are now captured as part of the ModelResponse and persisted in state.
⚡ No significant performance impact expected.

---

🧪 How to Test

• Run tasks/examples/llm_node_tool_simulation. Observe if tool_calls in LLM Node are being captured in the output data.
• Add graph_post_process in graph_config to run Graph Post processing job. Observe a new processed file with prefic as processor name being stored along with output file.

---

📸 Screenshots

N/A

---

✅ Checklist

• Lint fixes and unit tests completed
• End-to-end task testing verified
• Documentation updated

---

🗒️ Notes

This update lays the foundation for agentic model evaluation and structured response pipelines in the LLM Node, making it easier to benchmark tool-augmented models in complex workflows.

[amitsnow]

LGTM 🚀