Prefer GPU for Demucs, migrate config, and surface worker device logs

[wingedonezero]

Motivation

• Demucs runs were falling back to CPU even when a GPU was detected, so the app should prefer GPU by default and preserve that behavior for existing users.
• Existing configs that previously forced cpu should be migrated to the new auto default so users benefit from available GPUs without manual changes.
• Subprocess behavior was opaque on success, making it hard to confirm which device Demucs actually used.

Description

• Change default and validation: set source_separation_device default to 'auto' and extend validation to accept auto and rocm in vsg_core/config.py, and migrate existing saved 'cpu' values to 'auto' when loading settings.
• Subprocess argument mapping: pass an explicit device preference into the Demucs worker from vsg_core/analysis/source_separation.py and map rocm/cuda preferences to the worker as cuda where appropriate.
• Worker device selection: update the worker (run_separation) to honor cpu, mps, cuda/rocm and auto preferences and to choose the best available device with safe fallbacks to CPU.
• Visibility: have the worker print the chosen device to stderr and surface all stderr lines in the parent process logs so successful runs show which device was used.

Testing

• No automated tests were run for these changes.

---

Codex Task