Standardize _get_scores() return type across all query strategies

[CoolJosh0221]

Standardize _get_scores() to return (np.ndarray, np.ndarray) across all strategies

Motivation

_get_scores() had inconsistent return types across strategies — zip iterators, list(zip(...)), or not implemented at all. This blocks future plans for supporting batch-mode querying and ALBL soft-advice vectors, which need a uniform interface to consume scores from any strategy.

Note: "Soft-advice vectors" = continuous score distributions over the unlabeled pool (vs. the current one-hot hard advice from make_query()), enabling smoother blending in ALBL's Exp4.P bandit.

Changes

Every _get_scores() now returns (entry_ids: np.ndarray, scores: np.ndarray) where higher = more informative. Empty pools return (empty_int_array, empty_float_array). The base class default raises NotImplementedError.

• 6 strategies updated to new return format (US, BALD, CoreSet, EpsilonUS, InformationDensity, DensityWeightedMeta)
• 3 strategies added new _get_scores() implementations (QBC, QUIRE, RandomSampling)
• 1 strategy extracted scores from C extension (HintSVM)
• 2 strategies raise NotImplementedError (VarianceReduction, ALBL)

Backward compatibility

No public API changes. make_query() still returns int. return_score=True still returns (ask_id, list(zip(...))).

Tests

24 new contract tests in test_get_scores.py covering return types, empty pools, and return_score compatibility. 3 existing test files updated for the new tuple format. 90/90 pass (11 pre-existing sklearn 1.8 multi_class failures are unrelated).