fix(outlook msg): strip NUL terminators from UTF-16 string properties

[fei (feiiiiii5)]

Problem

A .msg property stored as PT_UNICODE (__substg1.0_*001F) can be terminated with a NUL code unit. _get_stream_data read the stream and then did data.decode("utf-16-le").strip() — and str.strip() does not remove U+0000. The terminator therefore survived into the converted Markdown:

terminator=b'\x00\x00'  ->  title repr: 'Confirmación de la reunión\x00'

Worse when the terminator is a single NUL byte: the buffer becomes odd-length, utf-16-le raises, and the existing fallback re-decodes the shifted bytes as UTF-8 with errors ignored, so non-ASCII text is silently destroyed — the printed repro loses the ó in Confirmación.

This premise is already accepted in the same file: the 8-bit helper at _outlook_msg_converter.py:281-283 documents that "some writers include trailing NUL terminators" and strips them, and merged #2295 applied that fix to the 001E path only. The 001F path was left behind.

Change

Trim the padding in _get_stream_data before decoding, one UTF-16 code unit at a time: drop a single odd trailing byte first, then b"\x00\x00" pairs. Removing pairs (rather than a whole run of NUL bytes) is what keeps the last real character intact — a lone 0x00 byte belongs to the preceding character and must not be eaten on its own.

Testing

New packages/markitdown/tests/test_outlook_msg_unicode_terminators.py, reusing the _FakeOleFileIO harness style already used by the msg tests: empty-string terminator variants for a populated property and for an empty property.

Base control — src/markitdown/converters/_outlook_msg_converter.py restored from 945314a45ddbe02935f2fd287b797dc0ba4a01e4 with git diff on that file printing nothing, so the run measures upstream code with the new tests kept:

$ python -m pytest tests/test_outlook_msg_unicode_terminators.py -q
5 failed, 1 passed in 0.18s
FAILED ...::test_unicode_terminators_are_removed[\x00]
FAILED ...::test_unicode_terminators_are_removed[\x00\x00]
FAILED ...::test_unicode_terminators_are_removed[\x00\x00\x00]
FAILED ...::test_empty_unicode_properties_are_omitted[\x00]
FAILED ...::test_empty_unicode_properties_are_omitted[\x00\x00]

With the patch applied, from packages/markitdown with PYTHONPATH pointing at this tree's src:

$ python -m pytest tests/test_outlook_msg_unicode_terminators.py -q
6 passed in 0.30s

No regression in the msg-related tests, same selection at both revisions:

base:  5 failed, 44 passed, 826 deselected, 8 errors   (pytest tests -k "msg or outlook")
head:  0 failed, 49 passed, 826 deselected, 8 errors

The 8 collection errors are the same files at base and head (test_pptx_svg.py, test_xlsx_images.py, ...) and come from optional extras missing in my environment, not from this diff — nothing in tests/ that imports the msg converter is among them.

Formatting: black --check with the repo's pinned 23.7.0 from .pre-commit-config.yaml → 2 files would be left unchanged.

Blast radius: _get_stream_data has exactly one call site (:252, the %s001F string property), so no attachment or other binary stream passes through the new trim.