chore: release 0.4.1

[maish]

Patch release: fix spanning body cells duplicated across columns on multi-page merge.

Fix

Docling represents a col_span=N cell by repeating its text across every column it covers. The merge round-trip (grid → DataFrame → grid) flattened those into N separate col_span=1 cells — duplicating a full-width description into every value column and displacing the real rate/cap values. A repeated multi-row header (col_span=6 banner) flowing in as a body row behaved the same way.

_dataframe_to_docling_data now matches each merged row back to its source grid row (across all member fragments) and re-emits the original spans. Rows the merger genuinely transformed (stitched continuations, folded overflow) fall back to the previous flat 1x1 rebuild. The match keys on original span metadata, never value equality, so coincidentally-equal adjacent values (two plan columns sharing a cap) stay separate cells.

No changes to the extractor or the merger.

Tests

• 106 passing (2 new regression tests: span preservation + the no-fuse guard for coincidental equals).
• Verified against a real multi-page insurance-benefit document: col_span=7 descriptions and col_span=6 repeated headers now preserved; rate/cap land in the correct plan columns.

🤖 Generated with Claude Code