Fix Linux CPU sidecar bundling CUDA, add cleanup workflow

Linux CPU sidecar: PyPI's default torch on Linux includes CUDA
(~800MB). UV_NO_SOURCES only bypasses our custom CUDA index but
still gets CUDA-enabled torch from PyPI. Now explicitly installs
CPU-only torch from pytorch.org/whl/cpu after sync. Same fix
applied to Windows.

New cleanup-releases.yml workflow (manual trigger):
- Configurable: keep N app releases, keep N sidecar releases
- Dry run mode (default) shows what would be deleted without deleting
- Protects v1.4.0 (last pre-Tauri release)
- Shows release sizes in MB

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

.gitea/workflows/build-sidecar-linux.yml

@@ -41,10 +41,11 @@ jobs:
           sudo apt-get install -y portaudio19-dev
 
       - name: Build sidecar (CPU)
-        env:
-          UV_NO_SOURCES: "1"
         run: |
-          uv sync
+          uv sync --no-sources
+          # PyPI's default torch on Linux includes CUDA (~800MB).
+          # Replace with CPU-only torch from the dedicated index.
+          uv pip install torch torchaudio --index-url https://download.pytorch.org/whl/cpu --force-reinstall
           .venv/bin/pyinstaller local-transcription-headless.spec
 
       - name: Package sidecar (CPU)