Preserve literal NA values and blank cells in Excel conversions

[Kunpeng Xie (pentaoa)]

Fixes #2498.

An Excel row containing NA, NULL, and a blank cell currently becomes | NaN | NaN | NaN |. Disable pandas' default NA-token recognition so literal cell text is preserved and blanks remain empty. Apply the same setting to XLS, XLSX, and the legacy showZeroes repair path.

The regression runs real workbooks through MarkItDown.convert_stream, including a small XLS fixture to avoid adding a workbook-writing test dependency. It checks several NA-like strings, blank cells, and integer values against the resulting Markdown table.

Validation:

• All three new format/path cases fail before the fix and pass afterward.
• Five focused Excel and legacy-repair tests pass.
• Eight existing XLS/XLSX file, hinted-stream, unhinted-stream, and file-URI conversion tests pass.
• Black and git diff --check pass. The full multi-format suite was not run; this change adds no dependencies.

AI assistance: OpenAI Codex was used for investigation, implementation, and local test execution.

[Kunpeng Xie (pentaoa)]

@microsoft-github-policy-service agree