fix(xlsx): tolerate legacy showZeroes sheet views

[he-yufeng]

Fixes #2063.

Some XLSX files still contain the worksheet view attribute name showZeroes. openpyxl 3.1+ expects showZeros, so loading those files raises a TypeError before MarkItDown can read the workbook.

This keeps the normal pandas/openpyxl path unchanged. If that path fails with the known showZeroes TypeError, MarkItDown rewrites worksheet XML entries in memory from showZeroes to showZeros and retries the read. The fallback is scoped to worksheet XML files and only runs for that specific compatibility error.

Validation:

• .venv\Scripts\python.exe -m pytest packages\markitdown\tests\test_module_misc.py -q -k "xlsx_legacy_show_zeroes"
• .venv\Scripts\python.exe -m pytest packages\markitdown\tests\test_module_vectors.py::test_convert_local -q
• .venv\Scripts\python.exe -m py_compile packages\markitdown\src\markitdown\converters_xlsx_converter.py packages\markitdown\tests\test_module_misc.py
• .venv\Scripts\python.exe -m ruff check packages\markitdown\src\markitdown\converters_xlsx_converter.py packages\markitdown\tests\test_module_misc.py
• git diff --check

[noezhiya-dot]

Clean fix for the showZeroes/showZeros compatibility issue in openpyxl 3.1+.

The approach is solid:

• Normal pandas/openpyxl read path stays unchanged (no performance impact)
• Fallback only triggers on the specific TypeError containing 'showZeroes'
• XML repair is scoped to xl/worksheets/*.xml files (doesn't touch other parts of the ZIP)
• In-memory BytesIO repair avoids writing temp files to disk
• The byte-level replace is safe here because showZeroes/showZeros is an attribute name that won't collide with cell data

The regression test is well-constructed — it builds a real XLSX, injects the legacy attribute, and verifies the full conversion pipeline recovers correctly.

One minor consideration: the string check is case-sensitive, which is correct since Python attribute names are case-sensitive. If openpyxl ever changes the error message format, this would silently stop triggering the fallback, but that's an acceptable trade-off.

LGTM.

[noezhiya-dot]

Clean fix for the showZeroes/showZeros compatibility issue in openpyxl 3.1+. The normal read path stays unchanged. Fallback only triggers on the specific TypeError. XML repair is scoped to worksheet files only. In-memory BytesIO repair avoids temp files. Regression test is well-constructed. LGTM.