fix(pptx): preserve masters/layouts/theme/fonts/bg + EMF fallback on save

[karthikmudunuri]

Summary

Three stacked regressions caused client decks to lose chunks of visual content after a single text edit. Verified end-to-end against Dickinson_Sample_Slides.pptx and eon-deck.pptx.

1. Slide masters, layouts, theme, embedded fonts dropped on save

pptxgenjs regenerates its own ppt/slideMasters/, ppt/slideLayouts/, ppt/theme/, and never emits ppt/fonts/. Anything that lived on the master / layout — backgrounds, brand bars, gradients, page numbers, footers, theme palettes, embedded brand fonts — disappeared.

preserveUnknowns now copies the source's chrome (also notesMasters, handoutMasters, tags) into the generated zip, splices the source's <p:sldMasterIdLst> / <p:notesMasterIdLst> / <p:embeddedFontLst> into presentation.xml, rewrites presentation.xml.rels and each slide's rels to point at the original layouts, updates [Content_Types].xml, and copies referenced media (renamed on collision with pptxgenjs's own media). Bails when source/output slide-size aspect ratios differ so 4:3 sources don't get their chrome stretched onto a 16:9 canvas.

2. Slide-level <p:bg> collapsed to flat hex

pptxgenjs's slide.background only emits a <a:solidFill> with a hex color, so theme refs / gradients / image-fill backgrounds were collapsing (e.g. <a:schemeClr val=\"tx1\"/> became <a:srgbClr val=\"151515\"/>).

A new per-slide pass copies the source slide's <p:bg> element verbatim, rewriting r:embed / r:link for image-fill backgrounds (with media copy) using the same machinery as injectIntoSlide. When the source slide has no explicit bg, the pptxgenjs flat-hex stand-in is stripped so layout / master inheritance can do its job.

3. EMF/WMF decode failures wiped the entire slide

parsePic returned null on metafile decode failure. Combined with upstream catches, a single un-decodable <p:pic> could wipe every other element on the same slide (Dickinson sample slides 2, 3, 9 — title + subtitle + logo all gone after one text edit on a different slide). The fallback now returns an UnknownElement so the source <p:pic> is re-injected verbatim and the EMF reference survives for PowerPoint to render natively.

Validation

Run on the two real attached decks (.context/attachments/):

Deck	Before	After
Dickinson_Sample_Slides.pptx (4:3)	slides 2/3/9 had empty spTree; 5/9 broken	9/9 slides retain content; slide 2's <a:schemeClr val=\"tx1\"/> theme bg survives
eon-deck.pptx (16:9)	1 layout, 0 fonts, 1 theme on save	28 layouts, 5 embedded fonts, 3 themes preserved

Changeset

Minor bump on @textcortex/slidewise — see .changeset/preserve-chrome-and-emf-fallback.md.

Test plan

• pnpm typecheck clean
• pnpm --filter @textcortex/slidewise test — 37/37 passing (2 new fixture-driven tests in chrome-preservation.test.ts)
• End-to-end import → save against both attached decks; saved files at /tmp/*.saved.pptx if you want to eyeball in PowerPoint before merge
• Reviewer: open the two saved files in PowerPoint and confirm Dickinson slide 2's black theme bg + eon-deck's brand fonts/layouts render as expected

Known not-yet-fixed

• 4:3 chrome preservation — Dickinson masters are still skipped due to the aspect-ratio guard. The clean fix is to drive output <p:sldSz> from the source and inverse-fit editor coords to source-canvas space — bigger refactor that touches the importer's fit + pxToInches. Tracking as follow-up.