Add Linux CPU and NVIDIA desktop support

[ZMXJJ]

Summary

• Add Linux Tauri bundle resource configuration plus CPU/NVIDIA preparation scripts.
• Add Linux runtime requirement splits for CPU and NVIDIA builds.
• Update sidecar CUDA selection so Linux NVIDIA bundles set VOCA_REQUIRE_CUDA=1 from the packaged manifest while CPU bundles keep CPU fallback.
• Keep inference backend/device information inside the existing Settings service diagnostics Inference Device row instead of rendering a standalone backend card.
• Document Linux install/build flows.

Testing

• node --check ./scripts/prepare-linux-resources.mjs && node --check ./scripts/run-tauri-dev.mjs && node --check ./scripts/run-tauri-build.mjs
• npm run typecheck:web
• npm --prefix app run build
• cargo fmt --check
• TAURI_CONFIG='{"bundle":{"resources":[]}}' cargo check
• TAURI_CONFIG='{"bundle":{"resources":[]}}' cargo test sidecar::tests
• python3 -m py_compile python-service/app/services/voxcpm_bridge.py python-service/app/services/sensevoice_onnx_session.py
• git diff --check
• git submodule update --init --recursive VoxCPM
• npm run build:linux:cpu produced the release binary and .deb; AppImage failed at linuxdeploy.
• node ./scripts/run-tauri-build.mjs --bundles deb rebuilt the Linux release .deb after the Settings UI cleanup.
• Launched /workspace/desktop/src-tauri/target/release/voca-desktop with VOCA_DEV_BUNDLE_ROOT=/workspace/desktop/.bundle-resources-linux and verified the real Linux Tauri app window.

Artifacts

linux_desktop_no_separate_inference_backend.mp4
Linux desktop settings without separate inference backend card

Notes

• npm run lint:web still fails on existing lint issues outside the Linux support changes.
• AppImage packaging currently fails in this environment when Tauri runs linuxdeploy; the Debian package is generated at /workspace/desktop/src-tauri/target/release/bundle/deb/Voca_0.5.0_amd64.deb.

_{To show artifacts inline, enable in settings.}