fix(genai): improve tool config, response metadata, and structured output

[pawel-twardziak]

Improves @langchain/google-genai provider with tool config handling, response metadata, structured output defaults, and modernized tests.

Changes

Core (chat_models.ts, utils/common.ts)

• tool_choice: "none" fallback: When tool_choice is set to "none" but no tools are provided (e.g. after bindTools([], { tool_choice: "none" })), invocationParams now emits toolConfig: { functionCallingConfig: { mode: FunctionCallingMode.NONE } }. Previously, the "none" choice was silently ignored when options.tools was empty because convertToolsToGenAI was never called.
• response_metadata.model: Both mapGenerateContentResultToChatResult (invoke) and convertResponseContentToChatGenerationChunk (stream) now include the model name in response_metadata, matching the pattern used by other LangChain providers (OpenAI, Anthropic). model_provider: "google-genai" was already present for streaming and is now also added to the non-streaming path.
• Default withStructuredOutput method: The method parameter now defaults to "functionCalling" instead of undefined, making the behavior explicit and consistent with other providers.
• LangSmith ls_structured_output_format metadata: withStructuredOutput now attaches ls_structured_output_format (with kwargs.method and schema) to the config for both "functionCalling" and "jsonSchema" paths, enabling LangSmith to track which structured output strategy was used.

Tests

• New unit tests for invocationParams with tool_choice: "none" (with and without tools).
• New integration tests for tool_choice: "none" via bindTools (both invoke and stream).
• Model upgrades: Updated all test model references from gemini-2.0-flash / gemini-1.5-pro / gemini-1.5-flash to gemini-2.5-flash.
• Embedding tests: Added explicit model: "gemini-embedding-001" to all embedding integration tests.
• GoogleSearchRetrievalTool tests: Skipped (describe.skip) - google_search_retrieval is no longer supported by current models; newer models use google_search instead.
• Context caching tests: Wrapped in describe.skipIf(!videoFileExists) so the suite is skipped gracefully when the large video fixture is not downloaded.
• Token count assertions: Relaxed toBe to toBeGreaterThanOrEqual for total_tokens checks and increased maxOutputTokens from 10 to 256 to avoid truncation flakiness.
• ESLint fixes: Replaced as any casts in invocationParams test calls with as Partial<GoogleGenerativeAIChatCallOptions>.
• Import fix: Changed tool_call_conversion.test.ts from @langchain/google-genai (package self-reference) to relative ../chat_models.js.

Related to #10432
This PR fixes the same problem, but for `@langchain/google-genai

---

Changes have been verified with the current state of the official docs for Google AI

---

integration tests

90% of the tests were not valid any longer; all fixed

unit tests

[changeset-bot]

🦋 Changeset detected

Latest commit: af1aafd

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 6 packages

Name	Type
@langchain/google-genai	Patch
@langchain/google-common	Patch
@langchain/google-gauth	Patch
@langchain/google-webauth	Patch
@langchain/google-vertexai	Patch
@langchain/google-vertexai-web	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR