fix(evaluation): include grounding metadata in hallucinations judge context

[chelsealong]

Please ensure you have read the contribution guide before creating a pull request.

Link to Issue or Description of Change

1. Link to an existing issue (if applicable):

• Closes: hallucinations_v1: sentence validator ignores grounding_metadata #7157

Problem:
HallucinationsV1Evaluator._create_context_for_step (src/google/adk/evaluation/hallucinations_v1.py) never included model-supplied grounding metadata (InvocationEvent.grounding_metadata) in the context string it builds for the sentence-validator judge — it only assembled developer instructions, user prompt, tool declarations, and per-event function_call/function_response/text parts. For agents that use built-in/model-internal grounding tools (e.g. VertexAiSearchTool, google_search), search results only ever arrive as groundingMetadata on the event, never as a function_response part. As a result, sentences that were genuinely grounded by search had no supporting evidence in the judge's context, and the judge labeled them unsupported, inflating the hallucination rate for any search-grounded agent. The sibling metric rubric_based_final_response_quality_v1.py already handles this correctly via get_grounding_metadata_as_json_str (added in #5834 for #5831), but hallucinations_v1.py was not updated at the time.

Solution:
Apply the same pattern used by rubric_based_final_response_quality_v1.py: call get_grounding_metadata_as_json_str (from llm_as_judge_utils.py) on the accumulated events for the current step and inject the result as a Grounding metadata: block in the context string, alongside the existing Tool definitions: block. Also added a short instruction to the sentence-validator prompt clarifying that Grounding metadata is trusted evidence from model-internal tools whose results may not otherwise appear in tool_outputs.

Testing Plan

Unit Tests:

• I have added or updated unit tests for my change.
• All unit tests pass locally.

$ python -m pytest tests/unittests/evaluation/test_hallucinations_v1.py tests/unittests/evaluation/test_llm_as_judge_utils.py tests/unittests/evaluation/test_rubric_based_final_response_quality_v1.py -q
........................................................                 [100%]
56 passed in 1.12s

Added TestCreateContext::test_create_context_for_step_includes_grounding_metadata, which builds an InvocationEvent with grounding_metadata set (and no content) and asserts the resulting context contains a Grounding metadata: block with the serialized web_search_queries. Confirmed this test (and 6 pre-existing ones whose expected context strings needed the new block) fails without the fix:

$ git checkout HEAD~1 -- src/google/adk/evaluation/hallucinations_v1.py
$ python -m pytest tests/unittests/evaluation/test_hallucinations_v1.py -q
...
FAILED tests/unittests/evaluation/test_hallucinations_v1.py::TestCreateContext::test_create_context_for_intermediate_step
FAILED tests/unittests/evaluation/test_hallucinations_v1.py::TestCreateContext::test_create_context_for_final_step
FAILED tests/unittests/evaluation/test_hallucinations_v1.py::TestCreateContext::test_create_context_for_step_includes_grounding_metadata - IndexError: list index out of range
FAILED tests/unittests/evaluation/test_hallucinations_v1.py::TestEvaluateInvocationsAgentTree::test_evaluate_invocations_multi_agents
FAILED tests/unittests/evaluation/test_hallucinations_v1.py::TestEvaluateInvocationsAgentTree::test_evaluate_invocations_agent_tree_skip_intermediate
FAILED tests/unittests/evaluation/test_hallucinations_v1.py::TestEvaluateInvocationsTimeWeather::test_evaluate_invocations_time_weather
FAILED tests/unittests/evaluation/test_hallucinations_v1.py::TestEvaluateInvocationsTimeWeather::test_evaluate_invocations_time_weather_skip_intermediate
7 failed, 14 passed in 1.30s
$ git checkout HEAD -- src/google/adk/evaluation/hallucinations_v1.py   # fix restored

Also ran the broader evaluation test directory (excluding files that error on collection due to optional gcp/docs extras — pandas, vertexai, etc. — not installed in this sandbox; unrelated to this change): 567 passed.

Lint (ruff, isort, pyink --check) clean on both changed files.

Manual End-to-End (E2E) Tests:

Not run — this is a prompt/context-construction change in an eval metric with no UI surface; covered by unit tests above.

Checklist

• I have read the CONTRIBUTING.md document.
• I have performed a self-review of my own code.
• I have commented my code, particularly in hard-to-understand areas.
• I have added tests that prove my fix is effective or that my feature works.
• New and existing unit tests pass locally with my changes.
• I have manually tested my changes end-to-end.
• Any dependent changes have been merged and published in downstream modules.

Additional context

This PR was prepared with AI assistance (Claude Code), per this repo's AI-Assisted Development guidance. All changes were reviewed and verified by running the test suite and lint locally before pushing.