MarkItDown 2Ways v0.3.0

MarkItDown 2Ways v0.3.0 adds the first conservative XLSX bidirectional round-trip tranche and a typed capability/diagnostic kernel.

Highlights:

• deterministic writable/read-only/derived capability decisions and coverage reports
• safe update_sheet_cells typed edits for existing scalar XLSX cells
• source-preserving SpreadsheetML reader, sparse patch writer, semantic readback, and native/package preservation verification
• identity Markdown round trips only for lossless simple string-cell regions
• no-op XLSX writes remain byte-identical
• differential output validation with openpyxl plus a real XLSX corpus smoke test
• hardened workbook/sheet authority: unique sheet IDs, relationship IDs, and worksheet targets
• bounded merged-range processing and sparse-grid materialization to fail closed on resource-amplifying inputs
• conflicting native value containers fail closed and patched value carriers preserve native XML ordering

XLSX tranche-one boundary:

• formula cells are read/preserved but read-only
• merged and rich-text cells are read-only
• row/column/sheet structural edits are unsupported
• styles, charts, drawings, formulas, merges, and other unsupported native structures are preserved rather than rebuilt
• this is not a claim of full Excel editing or full XLSX feature parity

Validation before merge:

• Python 3.10, 3.11, 3.12, and 3.13 package suites: green
• 804 passed, 33 skipped on the final tested merge tree
• OCR matrix Python 3.10–3.13: green
• pre-commit: green
• actual merge tree is byte-identical to the GitHub synthetic merge tree tested by CI

This is a fork-specific GitHub release; it does not replace the upstream MarkItDown PyPI package identity.