Uniform kwargs for processors of audio-text models

[leloykun]

What does this PR do?

• Uniformizes kwargs for processors of audio-text models.
• An extension of https://github.com/huggingface/transformers/issues/31911
• NOTE: don't review nor merge until this PR is complete: https://github.com/huggingface/transformers/pull/32841

TODO Models:

• [x] Clap
• [x] CLVP
• [x] MusicGen Melody
• [ ] Qwen2 Audio
• [x] Seamless M4T
• [x] SpeechT5
• [x] Wav2Vec2 Bert

TODO tests

• [ ] Add audio-text-specific processor tests
• [ ] Remove unnecessary/duplicated tests

Models with special args (will not be done in this PR):

• PopPiano

Models with weird _in_target_context_manager logic (will not be done in this PR):

• MusicGen
• SpeechToText
• Wav2Vec2
• Wav2Vec2 w/ LM
• Whisper

Fixes # (issue)

• https://github.com/huggingface/transformers/issues/31911

Who can review?

@zucchini-nlp @molbap @yonigozlan

[leloykun]

@zucchini-nlp the tests are failing because of this: https://github.com/huggingface/transformers/pull/32921