fix(tools): use full query text and own logger in PreloadMemoryTool

[chelsealong]

Fixes #6884

Problem

PreloadMemoryTool.process_llm_request in preload_memory_tool.py had two issues, both present on main:

1. The search query was read only from user_content.parts[0].text. If the first content part is non-text (a file, image, or a placeholder text part some LLM providers require ahead of a non-text part), the tool either returns early and never searches memory, or searches memory using the placeholder text instead of the user's actual question. The memory-rendering side of the same file (_memory_entry_utils.extract_text) already joins text across all parts — only the query-building side didn't.
2. A failed search_memory call was logged via the root logging module instead of the module's own google_adk.* logger, so applications that configure logging by the google_adk namespace never see the failure. Combined with (1)'s silent early return, a memory backend that is down or misconfigured is indistinguishable from "no memories matched".

Fix

• Build the query by joining text across every part of user_content, mirroring the existing extract_text idiom, instead of only looking at parts[0].
• Log retrieval failures through the module's own logger (google_adk.<module>) instead of the root logging module.

This keeps the existing fail-open behavior (per #3069) — retrieval failures still don't raise — but they are now observable through the correct logger, and the query no longer silently drops or mis-targets when the first part isn't text.

Test plan

Added two tests to tests/unittests/tools/test_preload_memory_tool.py:

• test_preload_memory_uses_text_from_every_part — a leading empty-text part must not blank out the real query in a later part.
• test_preload_memory_logs_search_failure_on_own_logger — a retrieval failure must be observable on the google_adk logger.

Verified both fail without the fix (reverted preload_memory_tool.py to the pre-fix version and reran):

FAILED tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_uses_text_from_every_part - AssertionError: Expected search_memory to have been awaited once. Awaited 0 times.
FAILED tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_logs_search_failure_on_own_logger - assert False
2 failed, 4 passed in 0.85s

With the fix restored, the full file passes:

tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_keeps_system_prefix_stable PASSED
tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_stays_after_function_response_boundary PASSED
tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_does_not_change_cacheable_prefix_fingerprint PASSED
tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_search_failure_is_noop PASSED
tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_uses_text_from_every_part PASSED
tests/unittests/tools/test_preload_memory_tool.py::test_preload_memory_logs_search_failure_on_own_logger PASSED
6 passed in 0.70s

Also ran the broader tools suite filtered on preload (7 passed, no regressions), and formatted with isort/pyink per CONTRIBUTING.md.

AI assistance disclosure

This change was written with the assistance of an AI coding agent (Claude), with the diff reviewed and tests verified by the submitter before opening this PR.