An unused category in a pandas Categorical column is counted as a zero-row group and lowers the score #994

Where: faircode/profiler.py _dimension() (line 335) and _analyze_groups(), tests/test_profiler.py.

Current gap: value_counts() on a Categorical column lists every declared category, including ones with no rows (a Parquet file written from a categorical dtype does this). The profile then reports a group with count: 0, imbalance_ratio: None, the flag "a subgroup is effectively absent (0 rows)" and a lower score, although no row is missing. CSV, JSON and the web engine can never produce such a group, so the same data scores differently by file format. #975 added tests for how this situation is rendered; this is about whether it should happen at all.

Repro:

>>> import pandas as pd; from faircode import profile
>>> col = pd.Series(pd.Categorical(["M"] * 30 + ["F"] * 30, categories=["M", "F", "X"]))
>>> d = profile(pd.DataFrame({"sex": col}))["dimensions"][0]
>>> d["n_groups"], d["imbalance_ratio"], d["dimension_score"]
(3, None, 63)        # with a plain object column: (2, 1.0, 100)

Suggested approach: count only observed values (observed=True, or .astype(object) before value_counts()), keep the zero-row path for reference baselines only if it is wanted there, and add a Parquet round-trip test.