fix(evals): open data file with utf-8-sig to support BOM-encoded jsonl inputs

[Mukller]

Fixes

Fixes #3670.

_validate_and_load_data passes the raw data path string to pd.read_json(data, lines=True).
pd.read_json uses the system default encoding when opening files, which is UTF-8 on most
systems but does not strip a UTF-8 BOM (\xef\xbb\xbf). A JSONL file written with
utf-8-sig encoding (BOM-prefixed UTF-8) causes pd.read_json to emit ValueError: Expected object or value because the BOM sits in front of the first {.

Fix

Open the data file explicitly with encoding="utf-8-sig" and pass the resulting file object
to pd.read_json. Python's utf-8-sig codec:

• Strips the BOM when present (handles utf-8-sig clients).
• Reads plain UTF-8 files unchanged (backwards-compatible).

# Before:
initial_data_df = pd.read_json(data, lines=True)

# After:
with open(data, encoding="utf-8-sig") as _f:
    initial_data_df = pd.read_json(_f, lines=True)

The error message and the surrounding try/except are unchanged; the only difference is
how the file is opened.

Reproduction

import json, pathlib, tempfile
from promptflow.evals.evaluate import evaluate

# Create a BOM-prefixed JSONL file
with tempfile.NamedTemporaryFile(mode='w', encoding='utf-8-sig', suffix='.jsonl', delete=False) as f:
    json.dump({"question": "What is 2+2?"}, f)
    path = f.name

# Before fix: ValueError: Failed to load data from …
# After fix:  loads successfully

Closes #3670.

[Mukller]

Self-Review

Root Cause

pd.read_json(data, lines=True) opens the file using the default system encoding.
On all common systems that default is UTF-8, which does not strip a Byte Order Mark.
A BOM (\xef\xbb\xbf) at the start of the file makes the JSON parser see the BOM
as the first character of the first token, which is not valid JSON → ValueError.

utf-8-sig is a Python built-in codec that reads UTF-8 BOM as a signal to skip the
BOM, while reading plain UTF-8 identically. It is the standard fix for BOM-related
encoding issues.

Correctness

Input encoding	Before	After
UTF-8 (no BOM)	OK	OK (unchanged)
UTF-8 with BOM	ValueError	OK (BOM stripped)
Other encodings	ValueError	ValueError (same)

The fix is strictly more permissive: existing UTF-8 files are unaffected, and BOM-prefixed
files now load correctly.

Scope

Two lines changed in _validate_and_load_data. The error handling, the return value, and
all callers are unchanged.