Implement VitsSVC resume training / finetune feature

[viewfinder-annn]

✨ Description:

This PR addresses bugs in VitsSVC module, mainly related to:

1. implement resume training / finetune feature
2. fix bug for multiple acoustic content training

🚧 Related Issues:

#91

👨‍💻 Changes Proposed:

• Fix bug in models/svc/vits/vits.py, removed useless content_vec feature dependency.
• Refractor VitsSVCTrainer to detach from TTSTrainer

🧑‍🤝‍🧑 Who can review?

@RMSnow @lmxue

🛠 TODO

• Further clean up config
• Refractor VitsSVCTrainer to detach from TTSTrainer

✅ Checklist:

• Code has been reviewed
• Code complies with the project's code standards and best practices
• Code has passed all tests
• Code does not affect the normal use of existing features
• Code has been commented properly
• Documentation has been updated (if applicable)

[RMSnow]

1. Implement VitsSVCTrainer from the SVC's trainer as the base instead of the TTS ones
2. Refine the config json files. Move some common setting to a config/svc.json, and make the exp_config.json as user-friendly as possible

[viewfinder-annn]

Just fixed problem 1, now this PR doesn't affect tasks other than SVC.

[viewfinder-annn]

Now the default training data config is the same across DiffWaveNet and VitsSVC. The exp_config.json is now user-friendly with fewer exposed options. I'll refractor SVC preprocess to svc.json in another PR soon.

[lmxue]

LGTM.