feat(llm-observability): $ai_tools capture in Langchain

[Twixes]

Tools are part of the LLM input prompt, and we haven't been capturing them at all in LLM observability!
This PR adds capturing of tools as $ai_tools to the Langchain callback handler. Does not touch SDK wrappers.

[greptile-apps]

PR Summary

This PR enhances LLM observability in the Langchain callback handler by adding support for capturing tools as $ai_tools, which are important components of LLM input prompts that were previously not being tracked.

• Added tools field to GenerationMetadata class in posthog/ai/langchain/callbacks.py to store tool configurations provided to the model
• Implemented tool extraction from invocation parameters in _set_llm_metadata method to capture tool data during LLM calls
• Added $ai_tools property to event properties when capturing generation events for complete tool visibility
• Created new test case test_metadata_tools in posthog/test/ai/langchain/test_callbacks.py to verify proper tool data capture and retrieval
• Focused implementation only affects Langchain callback handler without modifying SDK wrappers

_{2 file(s) reviewed, 1 comment(s)
Edit PR Review Bot Settings | Greptile}

[greptile-apps]

logic: The 'required' array references 'query_description' and 'query_kind' parameters that don't exist in the properties object, which only has 'bar'.

Suggested change

	"required": ["query_description", "query_kind"],
	"required": ["bar"],