[Hanabi9248]

standardize_dataset collapses both singleton axes when extracting targets. For a one-row DataFrame with one target, y becomes a scalar and assigning its sample index raises AttributeError. With multiple targets it becomes a Series indexed by target names and assigning the one-row protected index raises a length mismatch.

Limit squeeze to the columns axis. Single-target results remain Series, multiple-target results remain DataFrames, and the sample axis is retained. This does not change target values, weights, column selection or missing-value policy.

Validation on main 3487791, using Python 3.11.16 with real pandas/NumPy/sklearn and complete package initializers (no mocks):

• Added regression: 13 methods. Baseline has 11 affected methods with 13 error events (including three dtype subcases), plus two passing zero-/multi-row controls. Candidate passes all 13.
• Cases cover target values/types, an external target Series, column/array weights, feature selection, multiple protected attributes, missing values and real LinearRegression.fit(**train._asdict()) handoff.
• The one-row multiple-target case uses the existing dropna=False path; other multiple-target filtering combinations are outside this fix.
• Exact source and added-test bytes checked; added-line whitespace checks pass. CPU1/768MiB/16 processes/private temporary files/single-threaded BLAS; peak whole-job memory 123449344 bytes; no dependencies installed.

Run: python tests/test_standardize_single_sample.py -v

The full repository, existing dataset suite, online dataset downloads, other Python/pandas versions and Linux were not run. This is independent of legacy BinaryLabelDataset #193 and the OpenMLStore refactor. AI assistance was used for preparation/review; the test results are actual local executions.