Fix JSON loader schema inference for files starting with a UTF-8 BOM (#8241) - #8243

Fixes #8241.

What's the bug

When a JSONL file starts with a UTF-8 BOM (\xef\xbb\xbf), load_dataset("json", ...) infers a different schema than the same file without the BOM.

The JSON builder's first-batch "mixed-struct-types" pre-scan parses each line with ujson to decide whether to preserve heterogeneous nested dicts as a Json type:

try:
    examples = [ujson_loads(line) for line in batch.splitlines()]
except ValueError:
    pass
else:
    json_field_paths += find_mixed_struct_types_field_paths(examples)

ujson rejects the leading BOM with ValueError, the broad except ValueError swallows it, and find_mixed_struct_types_field_paths is never called. PyArrow's read_json tolerates the BOM and instead unifies the schema, materializing spurious nulls for fields absent from individual records. So the user-visible content silently changes depending on whether the file happens to start with a BOM.

Fix

Strip a leading UTF-8 BOM from the first batch (guarded to the start of the file) so the pre-scan and PyArrow operate on the same bytes.

Test

Added a regression test: a BOM-prefixed variant of the existing jsonl_file_with_lists_of_dicts_of_varying_keys fixture, asserting it produces the same schema (EXPECTED_LISTS_OF_DICTS_WITH_VARYING_KEYS) as the non-BOM file. It fails on main and passes with this change.

ruff check, ruff format, and pytest tests/packaged_modules/test_json.py::test_json_generate_tables all pass locally.