Fix test_classification.py to reflect current classification.py

[Copilot]

test_classification.py had three mismatches with the actual classification.py implementation that caused import errors and incorrect test assertions.

Changes

• Remove stale import — from animl.split import get_animals references a module that no longer exists. Replaced the single usage with inline filtering: detections[detections["category_label"] == "animal"].reset_index(drop=True)

• Fix test_non_dataframe_detections_raises — used "not_a_dataframe" (a str), which classify() accepts as a single file path, so no error was raised at input validation. Changed to a dict input to actually hit the unhandled-type branch. Updated expected exception from ValueError to AssertionError to match the raise AssertionError("Input must be a data frame of crops, single file path or vector of file paths.") in the code.

• Fix test_empty_animals_with_empty_detections — single_classification reads empty['category_label'] to populate prediction; the test's empty DataFrame was missing this column. Replaced the pre-built prediction/confidence columns with category_label.