Fix LangChainToolAdapter pydantic schema error for tools with internal fields

[tysoncung]

Replaces PR #7241 (by TyCoding) — rebased on current main.

Summary

Some LangChain tools (e.g., GoogleDriveSearchTool) have an args_schema that includes internal fields like run_manager (CallbackManagerForToolRun) which pydantic cannot serialize. This causes a PydanticSchemaGenerationError when creating a LangChainToolAdapter.

Fix

Rebuild the pydantic model using only the filtered fields from the LangChain tool's .args property, which already excludes internal fields like run_manager and callbacks. This is consistent with how LangChain itself exposes tool arguments.

Related Issue

Fixes #6385

CC: @TyCoding (original author) @anandfresh (original reviewer)

[tysoncung]

👋 All 7 CI workflows are in status — needs a maintainer to approve the run since this is a fork PR. Could someone with write access approve the CI workflows? Changes pass GitGuardian and CLA already. Thanks! 🙏

[avinashkamat48]

This rebuilds the args model from schema_cls.model_fields, which only exists on Pydantic v2 models. LangChain still supports/tools may expose Pydantic v1 args_schema classes with __fields__, and the previous code accepted the schema object directly. For those tools this will raise AttributeError during adapter construction. Could you handle both v1 and v2 schema field APIs, or add a regression test for a LangChain tool whose args_schema is a pydantic.v1 BaseModel?