Use OpenClaw chat for image-to-3D

[iamlukethedev]

Summary

• replace the direct provider route with an OpenClaw gateway chat.send based image-to-3D workflow
• preprocess uploaded images inside Claw3D into a compact visual summary, then ask a temporary helper agent to return a structured low-poly recipe over chat
• keep local rendering and GLB export in Claw3D so the generated asset still matches the matte retro office style
• add focused unit coverage for the asset helpers, gateway prompt/recipe parsing, and multipart route handling

Testing

• npm run typecheck
• npm run test -- --run tests/unit/pictureAsset.test.ts tests/unit/pictureModelGeneration.test.ts tests/unit/pictureModelRoute.test.ts
• npx eslint src/features/retro-office/RetroOffice3D.tsx src/features/retro-office/core/types.ts src/features/retro-office/core/geometry.ts src/features/retro-office/core/pictureAsset.ts src/features/retro-office/objects/pictureProps.tsx src/lib/office/pictureModelGeneration.ts src/app/api/office/picture-model/route.ts tests/unit/pictureAsset.test.ts tests/unit/pictureModelGeneration.test.ts tests/unit/pictureModelRoute.test.ts
• npm run build

Notes

• this avoids introducing a second paid provider key into Claw3D by delegating generation through the existing OpenClaw runtime connection
• because this is a no-OpenClaw-code-change workaround, the model only sees the frontend-derived visual summary rather than a first-class image attachment through the gateway
• the existing non-blocking optional openclaw build warning remains unchanged