A column mixing the number 1 and the string "1" gives two groups with the same label in Python and one group in JS #993

Where: faircode/profiler.py _dimension() (col.dropna().value_counts(), line 335), assets/profiler-engine.js nunique() / the group counter, tests/test_json_edge_cases.py.

Current gap: value_counts() keeps the integer 1 and the string "1" apart, then _analyze_groups() stringifies the labels, so the report lists 1 twice. JS keys groups by their string form and shows one 1 group with the combined count. A JSON column that mixes types (common when one exporter quotes some numbers) is therefore profiled differently by the two engines, and the Python report has two rows a person cannot tell apart.

Repro:

$ python3 - <<'E'
import json, tempfile, os
from faircode.loaders_extra import read_table
from faircode import profile
p = os.path.join(tempfile.mkdtemp(), "m.json")
open(p, "w").write(json.dumps([{"race": v} for v in [1, "1", "a", "b"] * 10]))
print([(g["label"], g["count"]) for g in profile(read_table(p), {"race": "race"})["dimensions"][0]["groups"]])
E
[('1', 10), ('1', 10), ('a', 10), ('b', 10)]      # JS: [('1', 20), ('a', 10), ('b', 10)]

Suggested approach: convert the column to strings before counting (after dropping nulls), so both engines key on the same text, and add a mixed-type JSON file to the parity tests.