feat: introduce PurplePrint Studio identity

[PurplePrintAI]

Summary

• Rebrand the independent OpenScreen fork as PurplePrint Studio across the editor, capture entry surfaces, app menus, icons, installer name and localized copy.
• Apply GraphOS-family neutral surfaces and restrained violet accents while preserving functional recording/status colors and the existing light-mode preference.
• Keep the technical app ID, user-data profile, .openscreen project format, CLI command and fork update origin stable. Document upstream provenance and keep Design Coach/MCP integration out of this frontend identity change.
• Make fresh Windows worktree builds stage the pinned FFmpeg SDK reliably, without writing through SDK links or deleting an unexpected extraction path.
• Resolve subscription prerelease updates from the fork's published release list, and validate a clean Windows native build with a fork-pinned speech payload.
• Prevent overlapping native Windows capture starts after the countdown and show a localized Starting… HUD state until the helper acknowledges recording.

Release impact

• This is an unreleased branding change, based on fork/main. No GitHub Release has been published. The owner approved a local Windows installation for validation.
• The installed Windows NSIS candidate is version 1.10.0-subscription.11 and is unsigned. The owner approved local .10 and .11 upgrades after verified backups. .11 is not published; public publication requires separate approval.
• Clean .10 and .11 Windows packaging passed. The owner retried after one "already running" error and recorded/exported MP4 on .10; .11 addresses that startup gap. The owner confirmed recording, MP4 export and an "up to date" check in installed .11. A real auto-updater download/install requires a later published release. Real Mac record/edit/export and Developer ID signing/notarization are deferred to the Mac lane. Design Coach/MCP integration belongs to its separate workstream.

Validation

• npm ci with Node 22.22.1 / npm 10.9.4.
• npm test: 190 files, 2,257 passed, 4 skipped after the native-start guard.
• App and test TypeScript checks, Vite build, docs check and i18n check passed; Biome passed with 13 pre-existing warnings.
• The earlier .8 npm run build:win:fork passed from this worktree. A fresh local .9 compositor rebuild stopped because libclang.dll was missing. Windows clean package run 36362047448 rebuilt the .10 native helpers and unsigned NSIS installer on a runner with LLVM and passed packaged CLI, payload and checksum checks. Its downloaded installer SHA-256 is 44930cfaa4643ca2f9c52abc5291e5d5dff067066de45ccae9cde823a2461db9; PR checks passed 15/15 at that source head.
• Packaged Windows app launched with an isolated user-data directory; PurplePrint Studio title and fork update origin verified.
• Before the live upgrade, the installed .7 fork created a synthetic-media project in an isolated profile. The .8 package retained the clip, light theme and preferences and exported a decodable MP4.
• The owner-approved .8 installer reused the per-user uninstall registration, removed the old executable, created a product-named child directory under the prior custom install root, and replaced the Start-menu shortcut. A verified local backup preceded installation. Three live project files and the AI configuration retained their hashes; private project contents were not opened/exported for this check.
• The installed .8 executable loaded copies of all three real project documents (three assets, three clips, three accessible media paths) and opened the editor in a separate test profile; no titles or footage were emitted.
• Packaged Windows WGC recorded a generated QA window with microphone, system audio and camera off; the saved project reopened and exported a decodable MP4. This used automated renderer control. The owner separately confirmed that the installed .8 HUD opened the source selector, recorded six seconds and stopped with a physical mouse; Studio then opened the new project.
• The .8 recording had no audio stream and its automatic transcription showed "Failed to fetch". The captioning demuxer resolved its WASM URL outside packaged dist/wasm; .9 corrects that URL. Unit tests cover packaged and development URLs, and installed .9 retesting passed as described below.
• The .9 Windows package contains the WASM at dist/wasm/web-demuxer.wasm and the corrected URL in its compiled transcription chunk. A fresh local compositor rebuild was blocked by missing libclang.dll; the packaged compositor binary is byte-identical to the installed, previously validated .8 one. The WGC helper was rebuilt. The package remains unsigned.
• Owner-approved .9 installation reused the same per-user directory; registered/executable versions report .9, nine installed payload hashes match the package, and the Start-menu shortcut points to the installed executable. All eight checked project/recording/AI-config hashes match the verified pre-upgrade backup.
• On installed .9, the owner repeated a video-only recording, saw no fetch error, and exported MP4 successfully. The new project saved a no-audio transcription verdict. No private frame or project text was exported for verification.
• GitHub /releases/latest returned 404 because this fork publishes subscription.N as prereleases. .10 now selects the highest published version in the installed channel, excludes drafts and unrelated channels, and validates the fork release URL. The update-checker and self-updater tests passed; a real .10 download/install remains unverified.
• The pinned Windows speech runtime now comes from this fork's successful build run 36294782126 with checked archive and individual file hashes. The clean Windows workflow uploads validation-only artifacts; it does not publish a release.
• Owner-approved .10 installation reused the existing per-user folder with no additional nesting. Registered/executable versions report .10, the Start-menu shortcut is intact, and ten checked project/recording/AI-config hashes match the verified pre-upgrade backup. All seven installed speech runtime hashes match the pinned manifest. The installed ASAR contains the new release-list endpoint and no old fork /releases/latest endpoint. Computer-use was closed immediately after GUI installation and launch checks.
• The owner reported a first .10 attempt where the three-second countdown ended while native capture was still starting, then a second Record click received "already running". A subsequent retry recorded and exported MP4. .11 guards the pending start synchronously, disables the Record control and shows Starting…; native hook/HUD regression tests and a Vite build passed. Clean .11 Windows run 36365238689 passed. The locally installed candidate has SHA-256 9631bba365c0e2720cabc26014572c4cf2e437f409ceb51217e0519da8da5872, matches its validation manifest and latest.yml, and remains unsigned. All 15 PR checks passed at its source head 771d6d54.
• Owner-approved .11 installation reused the existing per-user folder after a verified backup of 2,868 profile and 188 installed files, plus the .10 rollback installer. Six checked project/recording/AI-config hashes remained unchanged; registered and ASAR versions report .11, the Start-menu target is intact, and seven installed speech binaries match the pinned manifest. Computer Use was reset immediately after installer launch checks. The owner confirmed capture/export and the installed-app update-check UI; the brief Starting… label was not separately observed.
• The current head 25f7366c differs from the installed build only in this changelog and the Studio identity document. Final clean Windows run 36369731004 and all 15 checks passed. Its validation-only installer has SHA-256 0deded77bbd039c85ef5700139d7f54ec4e638e04e74af573347e7cee4693e9f, matches its feed and manifest, and is unsigned; this documentation-only rebuild was not reinstalled. No GitHub Release has been published.
• macOS validation run 36358715037 built .9 Apple Silicon and Intel DMGs. Both passed architecture, native payload, bundle ID, ad-hoc signature structure, CLI boot and mounted-DMG checks. Downloaded DMGs matched their SHA-256 manifests; neither is notarized.

Review focus

• Visual identity and upstream attribution.
• Technical identity, profile continuity and updater behavior after the installed path transition.
• Windows native package reproducibility and prerelease channel selection.
• FFmpeg SDK staging behavior on Windows and Linux.