perf(ShaderView): share one GPUDevice across all ShaderViews

[blazejkustra]

What

Each ShaderView previously requested its own adapter + device. A screen that mounts several effects (Banners, card stacks, the home list's live thumbnails) therefore spun up N GPUDevices on the one background runtime — real GPU/memory cost and slower init.

This introduces a module-level shared GPUDevice singleton (getSharedGPUDevice) that every view awaits. The promise is cached for the JS runtime's lifetime and self-heals: if the device is lost, the cache is cleared so the next mount requests a fresh one.

Lifecycle change

Because the device is now shared, useWGPUSetup no longer destroys it on unmount. Per-view resources (the uniform buffer) are still torn down by the render loop; pipeline/shader modules are released when the loop's worklet closure is GC'd.

Verification (argent, iOS — iPhone 17)

• Home list renders all live shader thumbnails simultaneously off one device.
• Banners screen renders all 5 effects (fireworks, confetti, fire, coin geyser, storm) concurrently — previously 5 devices, now 1.
• Unmount → remount (Back, then re-open Banners) reuses the still-alive device with no blank canvas or crash.

Files

• src/utils/gpuDevice.ts (new) — the singleton
• src/hooks/useWGPUSetup.tsx — use the shared device; drop per-view device destroy