[Feature]: Export to editable PPTX (native PowerPoint objects, not flat images)

[TTigger]

Problem

open-slide can export to static HTML and print-to-PDF, but there's no PowerPoint export. For audiences and orgs that live in PowerPoint / Keynote / Google Slides, a deck currently has to be hand-rebuilt before it can be handed off or co-edited.

The usual shortcut — screenshot each slide into a .pptx — is what most AI slide tools do, and it produces a deck where every slide is one flat image: you can't fix a typo, re-theme, or nudge a box. There's no way today to hand someone a .pptx where each block (text, image, shape) is an independent, editable object, the way a hand-built deck is.

Proposed solution

Add an "Export as PPTX" item to the existing Download menu (next to HTML/PDF), fully client-side, that turns each page into a real PowerPoint slide where every block is a native object:

• text → native text boxes (per-run bold/italic/color/size/letter-spacing; <em>/<strong> become their own runs)
• <img> → native pictures (embedded as base64)
• filled / bordered / rounded boxes → native autoshapes
• <table> → native tables
• linear gradients → native vector gradient fills (recolorable in PowerPoint)
• only un-mappable decorative graphics (inline SVG, blurred/blended overlays) fall back to a rasterized image of just that minimal subtree — never the whole slide

Mechanics: mount each page offscreen at the fixed 1920×1080 canvas (reusing the existing PDF-export lifecycle — wait for fonts + animations to settle), walk the DOM, read geometry/style via getBoundingClientRect / getComputedStyle, and emit one PptxGenJS object per node. 1920px @ 96dpi maps exactly to a 20″×11.25″ 16:9 slide, so positioning is precise. Gradients are post-patched into the OOXML with the already-bundled fflate (PptxGenJS can't emit gradients itself).

API: exportSlideAsPptx(slide, slideId, onProgress?), exported from @open-slide/core, wired into the slide Download dropdown using the existing progress toast.

I've prototyped this end-to-end and verified it across the demo decks — happy to open a PR (will link here).

Alternatives considered

• Screenshot / PDF-rasterize each slide into a .pptx — simplest, but produces non-editable flat images (the exact thing we want to avoid).
• python-pptx / server-side conversion — needs a backend; open-slide's export is fully client-side today.
• Re-implementing layout into pptx from scratch — unnecessary: pages already render at a fixed 1920×1080 with absolute pixels, so we can read the resolved geometry directly instead of re-deriving layout.

Which package would this affect?

@open-slide/core (runtime, inspector, present mode, CLI)

Additional context

Dependencies: PptxGenJS (the only mature, browser-capable generator of editable pptx; its only dependency is JSZip) + html-to-image (raster fallback). Both are lazy-imported via dynamic import() — exactly like fflate in export-html.ts — so they stay out of the main bundle. The gradient post-patch reuses the existing fflate dependency.

Verified (structural, by unzipping the OOXML; all open in PowerPoint without a repair prompt):

Deck	Native editable objects	Images	Notes
smoke test (controlled)	100%	0	gradient → vector gradFill
raycast-api (9 slides; gradients/SVG/images/blur)	~88%	brand logos / blur only	165 native text runs
claude-code-intro (9 slides; full-page noise overlay)	~93%	1/slide (the noise overlay)	rest native

Known limitations (intentional / honest):

• Web fonts can't be embedded by PptxGenJS → substituted to system fonts (Aptos/Consolas/Georgia); text may re-wrap slightly.
• flex/grid is baked to absolute coordinates (editable, but not reflow-aware).
• radial/conic gradients fall back to a solid first stop (only linear is vectorized).
• inline SVG / blur / mix-blend decorative graphics are rasterized (minimal subtree, resolution-capped).
• transitions/animations are not exported — the final settled state is captured (same behaviour as the PDF export).