Update daily CI Docker image to torch 2.13.0 / CUDA 13.0

[ydshieh]

Summary

• Bumps the transformers-all-latest-gpu Docker image from torch 2.11.0 / CUDA 12.6 to torch 2.13.0 / CUDA 13.0
• Updates TORCHCODEC from 0.11.0 → 0.15.0 to match
• Pins torchaudio==2.11.0 for torch >= 2.11.0: torchaudio is in maintenance mode and 2.11 is its final release — it is compatible with torch 2.11+ and does not pin torch to a specific version
• Adds a warning comment explaining that torchaudio only publishes wheels for the CUDA version it was released with (e.g. 2.11.0 only has a cu130 build); installing torch with a different CUDA suffix would silently keep the mismatched torchaudio build and cause a RuntimeError at import time
• Adds explicit CUDA=cu126 build-arg to the existing torch 2.8.0 CI job in build-docker-images.yml so it continues to work now that the Dockerfile default changed

Notes

• Why cu130 and not cu132? torchaudio 2.11.0 only has a cu130 wheel — there is no cu132 build. Installing torch+cu132 alongside torchaudio+cu130 causes torchaudio._check_cuda_version() to raise a RuntimeError on import, which propagates through the transformers lazy-module system and surfaces as a misleading ModuleNotFoundError: Could not import module 'is_clearml_available'.

🤖 Generated with Claude Code

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 30825863917
Result: success | Grafana metrics are not available yet.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.