fix(exports): preserve SVG HTML labels in image exports

[yaacovcorcos]

Problem

Mermaid's sanitized, HTML-compatible SVG is not always valid standalone XML (notably HTML line breaks). Separately, loading an SVG containing HTML labels from a blob URL can taint Chromium's canvas and prevent PNG encoding. Copying a saved SVG in the shared image viewer follows the same affected path.

Targeted fix

• Serialize Mermaid output through an inert DOM into valid SVG/XML, preserving SVG, XHTML and MathML content, geometry, IDs and references.
• Share one image decoder: SVG uses a self-contained data URL in the browser's non-interactive image context; raster images retain short-lived blob URLs. No browser-security relaxation or remote renderer.
• Give the diagram raster copy explicit dimensions and preserve existing output bounds. Release canvases after encoding, including failures.
• Preserve original-file downloads, asset fetch/CORS checks, clipboard handoff, Mermaid source and on-screen rendering. No dependency, provider, guidance or UI changes.

The production change is confined to three files and removes duplicate decoding code. Most added lines are regression tests.

Verification

• Before realignment, full pnpm run test: 19,903 passed, 108 skipped across 27 package suites. The original PR CI passed in full.
• After rebasing onto current main and Mermaid 12.0.0: 70 focused tests passed; the existing Mermaid rendering smoke passed with zero XML/PNG failures; the deeper real-Electron/Chromium export test passed 28 diagram/theme cases.
• The Chromium check covers SVG structure/namespaces, actual label pixels, PNG copy/download consistency, copying saved SVGs, Unicode/RTL/line breaks, invalid XML, bounded images, 24 concurrent SVG decodes and PNG/JPEG copy. External-resource requests: zero.
• After realignment, frozen-lockfile install, formatting, lint, full type checks, full build and git diff --check passed. Existing unrelated lint warnings remain; no lockfile change.
• Added the export regression command to the existing CI Test job under Xvfb. Linux CI is pending.

No computer-use or manual visual acceptance was performed. A target-isolated app from the rebased commit is running for owner review. The targeted hidden Electron test uses a temporary profile and does not touch the user's clipboard.

Relationship to the other Mermaid work

Rebased onto current main after #276 merged. The sole command-list conflict retained both complementary smoke tests, and #276's rendering test now requires successful standalone SVG and PNG export instead of preserving the old known-failure baseline. No repair-UX or rendering behavior was changed.

This fixes export correctness; SVG readers still need HTML-label (foreignObject) support. Converting HTML labels to SVG text is not included.

Documentation impact: Updated — docs/internals/scient-chat-diagrams.md.

Model and harness: GPT-5; Codex desktop harness.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.5 KiB	13.6 KiB	+103 B (+0.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	+2 B (+0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.4 KiB	6.5 KiB	+101 B (+1.5%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	56.2 KiB	57.0 KiB	+822 B (+1.4%)	66.4 KiB	✅
Codex	Live turn messages	8	8	0 (0.0%)	21	✅
Claude	Total thread wire	13.6 KiB	13.5 KiB	−81 B (−0.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	+5 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−86 B (−1.3%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.8 KiB	57.0 KiB	−840 B (−1.4%)	66.4 KiB	✅
Claude	Live turn messages	8	8	0 (0.0%)	21	✅

Baseline: 7937489 · PR result: abde9e5 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 114.1 KiB
• Claude decoded thread snapshot: 114.7 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.