Problem

_TextFileDifferentiator._is_json calls json.load() directly on the raw file bytes when libmagic guesses a generic text/plain MIME type and the content starts with [/{:

try:
    with self._ctx.open() as file:
        json.load(file)
    return True
except json.JSONDecodeError:
    return False

If the bytes aren't valid UTF-8, json.load() raises UnicodeDecodeError (via its internal detect_encoding step) — a different exception type than json.JSONDecodeError, so it's not caught. It propagates straight out of the public detect_filetype() API instead of falling back to FileType.TXT, the behavior for any other content libmagic can't identify more specifically.

Repro

import io
from unstructured.file_utils.filetype import detect_filetype

data = b"{not json at all just braces \xe9 text} and more padding"
detect_filetype(file=io.BytesIO(data), metadata_file_path="filename.pdf")
# UnicodeDecodeError: 'utf-8' codec can't decode byte 0xe9 in position 29: invalid continuation byte

Fix

Broaden the except clause to also catch UnicodeDecodeError.

Testing

• Added test_it_does_not_crash_on_non_utf8_bracket_prefixed_text_with_no_extension, confirming detect_filetype() returns FileType.TXT instead of raising. Confirmed it fails against the pre-fix code and passes after.
• Ran the full test_unstructured/file_utils/ suite (380 passed, 1 xfailed — no regressions).
• ruff check/ruff format --check clean. Bumped unstructured/__version__.py and added a CHANGELOG.md entry per the contribution checklist.

Note: this bumps to the same 0.25.1 as my other open PR (#4397), both branched from the same base — whichever merges second will likely need a quick rebase on the version files, happy to handle that when the time comes.

Disclosure: I used an AI coding assistant (Claude) to help identify this bug and draft the fix. I independently reproduced the crash before writing the regression test and ran the full local test suite before opening this PR.