LangGraph simple example

[benjibc]

• Simple database example
• Chat example

[Copilot]

Pull Request Overview

This PR introduces LangGraph support to the eval_protocol library by adding a new rollout processor and example implementations. The changes enable evaluation of LangGraph-based applications through a dedicated processor that handles conversion between eval_protocol and LangChain message formats.

• Adds LangGraphRolloutProcessor for processing LangGraph-based evaluations
• Implements database query example using Chinook database with LangGraph
• Provides simple chat example with Fireworks model integration

Reviewed Changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
eval_protocol/pytest/default_langchain_rollout_processor.py	Complete rewrite to support LangGraph with message conversion and error handling
eval_protocol/adapters/langchain.py	Simplified message serialization and added EP to LC conversion function
tests/pytest/test_langgraph_processor.py	Unit tests for the new LangGraph processor functionality
tests/chinook/langgraph/test_langgraph_chinook.py	Integration test using Chinook database with LangGraph
tests/chinook/langgraph/graph.py	LangGraph implementation for database queries
examples/langgraph/test_langgraph_rollout.py	Example evaluation test using LangGraph
examples/langgraph/simple_graph.py	Simple LangGraph implementation with Fireworks integration
examples/langgraph/data/simple_prompts.jsonl	Test data for LangGraph examples
requirements-dev.txt	Added LangGraph and LangChain dependencies
eval_protocol/pytest/handle_persist_flow.py	Fixed dataset name generation for Fireworks API compatibility
eval_protocol/adapters/bigquery.py	Consolidated exception handling and removed unused import

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[dphuang2]

"provider" is a pydantic thing, is it necessary here?

[benjibc]

        model,
        model_provider=model_provider,
        temperature=temperature,
        reasoning_effort=reasoning_effort,
    )
```, it is a langgraph thing. 

[dphuang2]

reading up on langgraph to understand this—its confusing at first

[dphuang2]

do these need to be two separate things? why not just have an agent factory:

Callable[[RolloutProcessorConfig], Graph]

[xzrderek]

i think we need to track row.tools, i just made similar change for pydantic processor: #168