QVAC-19266 infra: migrate tts-ggml desktop CI to registry-based model download

[freddy311082]

🎯 What problem does this PR solve?

• Desktop integration CI for tts-ggml was installing Python + a heavy venv (Torch, librosa, numba, llvmlite) and converting HF weights to GGUF on every run via scripts/convert-models.sh. Slow (~10–30 min cold), fragile (numba wheels on macOS x64 / Python 3.12), and redundant: the same GGUFs are already published in the QVAC model registry.
• Mobile CI already downloads the GGUFs at runtime from the registry (PR feat(tts-ggml): Android dynamic ggml backends #2168), so desktop was the odd one out.
• Integration tests treated a missing GGUF as t.pass('Skipped: ...'), so a failed registry fetch or a misconfigured runner went green instead of failing the suite.

📝 How does it solve it?

• New CLI packages/tts-ggml/scripts/download-tts-ggml-models.js using @qvac/registry-client to pull the six GGUFs (chatterbox-t3-turbo, chatterbox-s3gen, chatterbox-t3-mtl, chatterbox-s3gen-mtl, supertonic, supertonic2) from S3 and rebadge the q4_0 filenames on-disk to the historical names the resolver expects.
• New npm script download-models:registry exposes the CLI.
• .github/workflows/integration-test-tts-ggml.yml: removed the Setup Python, Cache TTS GGML venv + GGUFs, and npm run setup-models steps. Replaced with the shared ./.github/actions/cache-models composite (gated to qvac-* runners, matching parakeet) + a Download GGUF models from registry step that runs npm run download-models:registry on cache miss.
• Converted 25 t.pass('Skipped: ... GGUF(s) not available') sites across six integration test files to t.fail(...) so a missing model is now a hard failure with a remediation hint (Run \npm run download-models:registry` or stage models locally.). t.pass('Skipped: reference audio missing')` is left in place — that asset lives in the repo, not the registry.
• Mobile workflow (integration-mobile-test-tts-ggml.yml) intentionally untouched: it already downloads at runtime.
• The Python conversion scripts (setup-venv.sh, convert-models.sh, scripts/convert-*-to-gguf.py) and their npm entries are kept for local-dev regeneration; they are simply no longer wired into CI.

🧪 How was it tested?

• npx standard clean on the new CLI and all six edited test files.
• YAML of the rewritten workflow parses OK.
• ReadLints clean on every edited file.
• CI manual dispatch to be run on the PR branch to validate the new download step end-to-end on every matrix entry.

[freddy311082]

CI here: https://github.com/tetherto/qvac/actions/runs/26484683916/job/77989511732

[github-actions]

Tier-based Approval Status

**PR Tier:** TIER1

**Current Status:** ✅ APPROVED

**Requirements:**
- 1 Team Member approval ❌ (0/1)
- 1 Team Lead OR Management approval ✅ (2/1)

**Bypass rule:** Triggered (2+ Team Lead approvals (Tier 1 exception)). This PR is approved regardless of tier.

---
*This comment is automatically updated when reviews change.*

[kinsta]

Preview deployments for qvac-docs-staging ⚡️

Status	Branch preview	Commit preview
✅ Ready	Visit preview	Visit preview

Commit: d238b20b7fb77a9bcfba73b90aa9dd5e1c770c26

Deployment ID: c0e49fe8-c32a-4c98-ac10-49fa3b1bbf46

Static site name: qvac-docs-staging-fazwv

[freddy311082]

CI ran here: https://github.com/tetherto/qvac/actions/runs/26756390277

[freddy311082]

/review