feat: add background noise export

[jbang2004]

Description

Add noise extraction functionality to speech enhancement models. This feature allows users to obtain both enhanced speech and extracted noise signals.

Changes

• Add extract_noise parameter to decode functions
• Modify model processing pipeline to support noise extraction
• Update ClearVoice interface to handle noise output
• Keep original functionality intact for backward compatibility

Implementation Details

• Added extract_noise boolean parameter (default: False)
• When extract_noise=True, functions return tuple of (enhanced_audio, noise_audio)
• When extract_noise=False, functions maintain original behavior
• Noise is calculated as: original_signal - enhanced_signal

Testing

• Tested with all speech enhancement models:
  • FRCRN_SE_16K
  • MossFormer2_SE_48K
  • MossFormerGAN_SE_16K
• Verified output quality and signal consistency

[zexupan]

Can you open a new branch for this? Thanks.