Bump PyTorch/Torchaudio to 2.10+, switch GPU wheels to CUDA 12.8, and update Python range

[EnisBerk]

Motivation

• Raise the minimum PyTorch/Torchaudio versions to address a reported security advisory and ensure users install a non-vulnerable PyTorch baseline (2.10+).
• Align the GPU pip index and requirements with PyTorch 2.10 wheels built for CUDA 12.8 and update environment/packaging Python guidance accordingly.
• Update installation and troubleshooting docs to reflect the new dependency and platform guidance.

Description

• Updated requirements_cpu.txt and renamed/updated requirements_gpu_cu128.txt to require torch>=2.10,<3, torchaudio>=2.10,<3, and torchvision>=0.25,<0.26 and to point the extra-index to the CUDA 12.8 wheel index.
• Reworked environment_cpu.yml and environment_gpu.yml to require python>=3.10,<3.15, pin pytorch/pytorch-gpu to >=2.10,<3, bump torchvision to >=0.25,<0.26, and install torchaudio>=2.10,<3 from the appropriate wheel index.
• Updated packaging metadata/comments in setup.py to reference requirements_gpu_cu128.txt and set python_requires='>=3.10, <3.15'.
• Updated docs (docs/installation.md, docs/common/troubleshooting.md, and README references) to mention CUDA 12.8 guidance, the new requirements filename, Python range, and torchaudio/FFmpeg compatibility notes.
• Replaced occurrences of the old CUDA 11.8 / cu118 references with CUDA 12.8 / cu128 where appropriate and added the torchaudio pin to the conda pip entries.

Testing

• Ran python -m py_compile setup.py to validate setup.py syntax, which succeeded.
• Validated environment_cpu.yml and environment_gpu.yml parse as YAML using yaml.safe_load, which succeeded.
• Parsed requirements_cpu.txt and requirements_gpu_cu128.txt to verify expected requirement entries were present, which succeeded.

---

Codex Task