fix: resolve Windows single-GPU training crash and matplotlib compati…

[unknown7751]

…bility

• Skip mp.Process subprocess for single-GPU runs; call run() directly. On Windows, PyTorch spawn-based child processes silently crash during CUDA context initialization, preventing any training from starting.
• Guard dist.init_process_group and DDP wrapping behind n_gpus > 1.
• Set num_workers=0 on Windows to avoid nested subprocess spawning. Remove persistent_workers and prefetch_factor which require num_workers > 0.
• Replace fig.canvas.tostring_rgb() with fig.canvas.buffer_rgba() in plot_spectrogram_to_numpy(). tostring_rgb() was removed in matplotlib 3.8.

Pull request checklist

• The PR has a proper title. Use Semantic Commit Messages. (No more branch-name title please)

• Make sure this is ready to be merged into the relevant branch. Please don't create a PR and let it hang for a few days.

• Ensure you can run the codes you submitted succesfully. These submissions will be prioritized for review:

  Introduce improvements in program execution speed;

  Introduce improvements in synthesis quality;

  Fix existing bugs reported by user feedback (or you met);

  Introduce more convenient user operations.

PR type

• Bug fix / new feature / synthesis quality improvement / program execution speed improvement

Description

• Describe what this pull request is for.
• What will it affect.

Screenshot

• Please include a screenshot if applicable