[open-swe] feat: Add LangSmith tool calling dataset and evaluation script

[open-swe]

Fixes #109

This PR adds a comprehensive script for setting up and evaluating LangSmith datasets for tool calling functionality.

What's Added

• python/openevals/tool_calling_dataset.py: Complete tool calling evaluation framework including:
  • Dataset creation function that formats examples for LangSmith
  • Tool call extraction functions that process various message formats
  • Evaluator that compares expected vs actual tool calls with detailed scoring
  • Integration with existing openevals utilities for seamless evaluation workflow

Key Features

• Handles multiple input formats (ChatCompletionMessage, BaseMessage, dict)
• Extracts tool calls with names and arguments from AI responses
• Provides accurate scoring based on exact matches of tool names and arguments
• Integrates with existing _run_evaluator and _normalize_to_openai_messages_list utilities
• Includes comprehensive error handling and edge case management

The script follows the existing openevals patterns and provides a complete solution for testing tool calling functionality with LangSmith datasets.