fix(flows): scope skip_summarization text-append to AgentTool

[lwangverizon]

fix(flows): scope skip_summarization text-append to AgentTool

Fixes #6230

Problem

PR #5974 (issue #3881) added a text part to the function-response event when
skip_summarization is True, so an AgentTool's output stays visible in UIs
that do not render function responses. The condition in __build_response_event
(src/google/adk/flows/llm_flows/functions.py) is not guarded by tool type, so
it fires for every tool that sets skip_summarization:

if (
    tool_context.actions.skip_summarization
    and 'error' not in function_result
    and has_displayable_result
):
    ...
    content.parts.append(types.Part.from_text(text=result_text))

Tools other than AgentTool often set skip_summarization for the opposite
reason: their function response is an internal acknowledgement (e.g. a
UI/widget-rendering tool returning {"status": "ok"}) that should not be
summarized or shown to the user. Force-converting that ack into a text Part
makes it bypass UI/SSE filters that only strip
functionResponse/functionCall/thought parts, so the raw payload is
surfaced to the user as visible text.

Issue #3881 and PR #5974 are both explicitly about AgentTool, so this is a
scope/regression issue rather than an intended behavior change for all tools.

Change

Guard the text-append with isinstance(tool, AgentTool):

if (
    isinstance(tool, AgentTool)
    and tool_context.actions.skip_summarization
    and 'error' not in function_result
    and has_displayable_result
):
    ...

• AgentTool keeps its text output (the The agent doesn't work as expected when "skip_summarization" is set to True in AgentTool #3881 behavior is preserved).
• Other tools that opt out of summarization no longer have their function
  response duplicated as a visible text part.
• AgentTool is imported lazily inside the function to avoid the
  agents -> flows -> tools -> agents circular import (the codebase already
  lazy-imports AgentTool elsewhere for the same reason).

Tests

• Adds test_skip_summarization_non_agent_tool_appends_no_text_part in
  tests/unittests/flows/llm_flows/test_functions_simple.py: a FunctionTool
  that sets skip_summarization=True and returns a dict produces only a
  function_response part and no text part.
• The existing AgentTool skip_summarization tests
  (tests/unittests/tools/test_agent_tool.py, added by feat(skills): Allow opt-in session state injection in skill instructions #5974) continue to pass,
  confirming the intended behavior is preserved.

Backwards compatibility

Narrows an over-broad behavior introduced in #5974 back to its stated scope
(AgentTool). No public API or signature change.