fix(optimization): treat explicit empty eval case ID lists as empty in LocalEvalSampler

[chelsealong]

Fixes #7168

Problem

LocalEvalSamplerConfig documents train_eval_case_ids and
validation_eval_case_ids as optional lists where all cases are
selected when a field is not provided. However,
LocalEvalSampler.__init__ used truthiness checks (or / if config.x:),
so an explicitly passed [] was indistinguishable from None and
silently expanded to every eval case in the set — turning a
"run zero cases" configuration into a full evaluation run.

Fix

Changed both checks in LocalEvalSampler.__init__ to is not None,
so only an omitted/None field triggers the "use all eval cases"
fallback. An explicitly empty list now stays empty, consistent with
the documented behavior and with sample_and_score(batch=[]).

# before
self._train_eval_case_ids = (
    self._config.train_eval_case_ids
    or self._get_eval_case_ids(self._train_eval_set)
)
...
if self._config.validation_eval_case_ids:
    ...

# after
self._train_eval_case_ids = (
    self._config.train_eval_case_ids
    if self._config.train_eval_case_ids is not None
    else self._get_eval_case_ids(self._train_eval_set)
)
...
if self._config.validation_eval_case_ids is not None:
    ...

Testing plan

Added two parametrized cases to the existing
test_local_eval_service_interface_init test in
tests/unittests/optimization/local_eval_sampler_test.py, covering an
explicit empty train_eval_case_ids and an explicit empty
validation_eval_case_ids.

Confirmed the new cases fail without the fix (reverted only the
source file via git checkout HEAD~1 -- src/google/adk/optimization/local_eval_sampler.py):

FAILED ...test_local_eval_service_interface_init[config_kwargs5-expected_attrs5] - AssertionError: assert ['train_set_1', 'train_set_2'] == []
FAILED ...test_local_eval_service_interface_init[config_kwargs6-expected_attrs6] - AssertionError: assert ['train_set_1', 'train_set_2'] == []
2 failed, 5 passed, 7 deselected

With the fix restored, the full test file and the rest of the
optimization suite pass:

$ pytest tests/unittests/optimization -q
35 passed, 12 warnings in 13.16s

Also ran isort --check-only and pyink --check on both changed
files — no issues.

AI assistance disclosure

This change was authored with the assistance of Claude Code (Anthropic).