Create a proper test to catch the multilabel behavior broken in 0.5.0

@afermg mentioned current test suite passed with the duckdb implementation but it failed to get the same result as in mAP notebooks.

Such use case(s) should be part of the test suite

Compare the duckdb implementation's current test behavior with the mAP notebooks, focusing on the multilabel use case that produced different results in version 0.5.0. Add a test covering that use case and confirm the test suite reproduces the expected notebook result.

Written by the indexing model from the issue text.