The web engine does not flatten nested JSON objects, so a nested column is dropped there but profiled in Python #992

Where: assets/profiler-engine.js recordsToTable() / parseJSON(), faircode/loaders_extra.py read_table() (the json_normalize call, #844), tests/test_json_edge_cases.py.

Current gap: the Python loader flattens a records-style JSON whose values are objects into dotted columns (loc.state), so {"loc": {"state": "TX"}} becomes a profiled column. The JS parser keeps the object as one cell, and since #973 such a column is skipped, so the same file gives different dimensions on the CLI and on the web.

Repro:

$ cat nested.json     # 30 records: {"sex": "m"|"f", "loc": {"state": "TX"|"CA"}}
$ python3 -c "from faircode.loaders_extra import read_table; print(list(read_table('nested.json').columns))"
['sex', 'loc.state']
$ node scripts/engine-js.js profile-json nested.json | python3 -c "import json,sys; print([d['name'] for d in json.load(sys.stdin)['dimensions']])"
['sex']

Suggested approach: flatten nested plain objects in recordsToTable() with the same dotted-name rule as pandas.json_normalize (lists stay as cells), and add the file above to the JSON parity tests.