fix(ai): capture tool calls in reasoning models

[skoob13]

Problem

LangChain passes original tool calls in additional_kwargs of a message for all models except for reasoning models, so they're not captured for OpenAI o* models.

Changes

Added tool calls for reasoning models by converting LangChain's shape to original OpenAI shape.

[greptile-apps]

Greptile Summary

This PR fixes a critical gap in tool call capture functionality for OpenAI reasoning models in the LangChain integration. Previously, tool calls were only being captured from additional_kwargs, but LangChain handles tool calls differently for reasoning models. The change adds a new conversion function _convert_lc_tool_calls_to_oai that transforms LangChain's tool call format into OpenAI's expected format, ensuring consistent tool call capture across all model types.

The changes include:

1. A new helper function to convert tool call formats
2. Updates to _convert_message_to_dict to handle tool calls in AIMessage objects
3. Comprehensive test coverage with test_convert_message_to_dict_tool_calls
4. Version bump from 6.2.1 to 6.2.2

Confidence score: 5/5

1. This PR is safe to merge as it fixes a missing functionality without disrupting existing behavior
2. The changes are well-tested, isolated to a specific feature, and follow a clear path for format conversion
3. Key files to review: callbacks.py - specifically the new conversion function and message dict handling

_{3 files reviewed, no comments}

_{Edit Code Review Bot Settings | Greptile}