fix(tools): allow transfer_to_agent to accept extra kwargs

[AlankritVerma01]

Fixes: #458

Background
When an LLM‐driven agent called transfer_to_agent with extra parameters (e.g. query, temperature), the tool raised:

TypeError: transfer_to_agent() got an unexpected keyword argument 'query'

because its signature only accepted agent_name and tool_context.

With this change, we:

• Update src/google/adk/tools/transfer_to_agent_tool.py to:

  def transfer_to_agent(
      agent_name: str,
      tool_context: ToolContext,
      **kwargs,
  ):
      """Transfer the question to another agent."""
      tool_context.actions.transfer_to_agent = agent_name

  so that any unexpected kwargs are silently ignored.
• Expand tests/integration/tools/test_transfer.py to include:
  1. A call with a single extra kwarg (query)
  2. A call with multiple extra kwargs (query, temperature, max_tokens)
     Each test is parametrized to run against both the GOOGLE_AI and VERTEX backends.

---

Test Plan

1. Run the targeted integration tests:

   pytest tests/integration/tools/test_transfer.py

2. Run the full suite to ensure no regressions:

   pytest

---

Checklist

• New integration tests added and passing
• Existing behavior unchanged for calls without extra kwargs
• Code follows repository style and linting rules
• References issue with Fixes: transfer_to_agent tool fails with unexpected kwarg #458

---

Next Steps / TODO
@hangfei @Jacksunwei
Per the existing TODO comment in transfer_to_agent_tool.py, we could further internalize this tool (e.g. rename to _transfer_to_agent or move into an internal/ submodule) so it doesn’t leak into the public API.
I can work on that. Let me know if I can / should?

[Jacksunwei]

Thanks for the PR!

But I think this is not the way to go. Model shouldn't be passing query to this tool. We need inspect the prompt and see why.

A better approach could be improving the docstring, which will be in the prompt.

[AlankritVerma01]

Thanks @Jacksunwei for the feedback! I’ve removed the **kwargs change and updated the docstring to clarify that only agent_name and tool_context are accepted and that parameters like query belong in the LLM prompt. Let me know if you’d like any further tweaks.

[Jacksunwei]

Let's remove math_agent to avoid confusing LLM.

[Jacksunwei]

tool_context is not for LLM use, so we don't want to mention it.

[Jacksunwei]

s/identifier/name.

[Jacksunwei]

We can just remove returns section, since there is none.

[Jacksunwei]

These are likely not useful to LLM, for prompt, it's better to be direct and concise.