faircode profile data.json silently misparses split-orient JSON that omits the index key, instead of raising or falling back

[yakew7]

Where: faircode/loaders_extra.py:36-51.

README.md (around line 928) documents that faircode profile data.json supports "JSON records (or split-orient)". The split-orient fallback is:

try:
    return pd.read_json(path)
except ValueError:
    return pd.read_json(path, orient="split")

The bug: for a minimal, natural split-format file that omits the optional index key — e.g. {"columns": ["sex","age"], "data": [["M",30],["F",25]]} (exactly the shape assets/profiler-engine.js's parseJSON() documents and accepts in its own comment: "2. Split format: { columns: [...], data: [...] }") — pd.read_json(path) (the default-orientation attempt) does not raise a ValueError, so the orient="split" fallback never runs.

Repro:

import pandas as pd, json, tempfile
data = {"columns": ["sex","age"], "data": [["M",30],["F",25]]}
# write to a .json temp file, then:
pd.read_json(path)
#   columns     data
# 0     sex  [M, 30]
# 1     age  [F, 25]

This returns a 2-row DataFrame with columns literally named ["columns", "data"]" and list-valued cells — silently wrong, no error, no warning — while the JS engine correctly parses the identical input into {columns:["sex","age"], rows:[{sex:"M",age:30},{sex:"F",age:25}]}`.

tests/test_loaders.py::test_read_table_json_orientations only covers DataFrame.to_json(orient="split") output, which happens to include an extra `"index"" key that does trigger the ValueError fallback — so this specific, documented-as-supported shape is untested.

Suggested fix: detect split-orient by checking for {"columns", "data"} (optionally "index") keys in the parsed JSON up front, rather than relying on pd.read_json's default-orientation guess to fail.