Release 0.6.0: preserving legacy edits

This increment extends existing-file edits rather than claiming complete legacy Office read/write parity. It is additive to the published 0.5.0 API.

Existing interfaces gain no required fields; DOC callable signatures gain only optional trailing limits. The default processing budgets intentionally refuse oversized inputs that previously could be accepted. Exhaustive TypeScript consumers of OleDocParagraphEditResult.reason must handle the new resource-limit and invalid-limits values. This behavior tightening and the new APIs justify the pre-1.0 minor version increment.

Historical ole-word-97.doc, sample-deck.ppt and picture-fixture.ppt were already tracked, unchanged, at upstream commit 240da4f07cb4aec2c0db39b1dcd043682d8270f5. Their origin remains unverified. The only new binary fixtures since that baseline are repository-owned synthetic doc/header-field.doc, ppt/native-text.ppt and xls/workbook-rich-strings.xls, each with its generator and SHA-pinned provenance. No new externally sourced or unlicensed binary was added. Fixtures, native output files and snapshots are excluded from the npm package.

• resizeCompoundFileStream preserves nested paths, unknown streams and raw directory metadata while growing/shrinking regular v3 streams. Ambiguous storage ownership and overlapping allocations fail explicitly. Mini resizing, mini/regular transitions, v4 resizing and DIFAT expansion remain unsupported.
• DOC paragraph growth uses the preserving resizer instead of reconstructing a flat container. Existing character-position feature guards remain enforced. Configurable main-text and piece-processing budgets bound malformed alias expansion before allocation/decoding.
• editXlsPreservedStringCell selects worksheet tabs and preserves old serialized SST/CONTINUE data, rich/phonetic strings, cell tables and formulas. New Unicode entries can span continuation records. Supported BOUNDSHEET/INDEX and ExtSST pointers are updated; unsupported pointer-bearing records are refused. Replacing a rich target creates plain text while preserving its XF and other cells sharing the original string. No formula recalculation is performed.
• New neutral synthetic PowerPoint and rich-string Excel fixtures have included Office generators and SHA-pinned provenance. No new external fixture is claimed licensed. Historical DOC/PPT provenance remains explicitly unverified.

Integrated validation passed 287 tests across 33 files, frozen installation, typecheck, build, independently installed package/API compatibility smoke, documentation generation and Pages checks. Nine native Word/Excel/PowerPoint16 comparisons passed: unchanged stream relocation in each format, nested DOC growth, Unicode and 24,000-unit SST edits, rich-string aliases and later-sheet selection, and ASCII/UTF-16 edits in the owned PowerPoint fixture. The integrated build regenerated all nine exact output hashes checked by native applications.

Native comparisons cover declared semantic fields and selected Excel character fonts. They do not establish rendered fidelity, widths, complete DOC runs, media/object behavior or all unsupported records. Applications opened explicit fixtures read-only with macros disabled; no personal documents or embedded objects were activated. Detailed local evidence records input/output/snapshot hashes, exact expected changes and these limitations.

Remaining work includes mini/v4 container resizing, richer DOC CP-table updates, more XLS record/cell conversions, full PPT models and synchronized mirrors, and VSD drawing models/writers. VSD/PUB support remains structural inspection and bounded metadata editing. Consumer integration belongs in coordinated follow-up work; no cross-repository source was changed here.