Loader downloads audio even when InputMode.TEXT_ONLY is specified

[cleanshafay]

When using the MMUSEDFallacy loader with InputMode.TEXT_ONLY, the loader still proceeds to download and process audio files, which is unexpected behavior. This causes unnecessary bandwidth and storage usage, especially in constrained environments.

---

Steps to Reproduce:

1. Install and import the dataset loader.

2. Use the following snippet to load the dataset in TEXT_ONLY mode:

   from pathlib import Path
   from mamkit.datasets import MMUSEDFallacy, InputMode
   
   base_data_path = Path('./data')
   loader = MMUSEDFallacy(
       task_name='afc',
       input_mode=InputMode.TEXT_ONLY,
       base_data_path=base_data_path
   )
   split_info = loader.get_splits('mm-argfallacy-2025')

3. Observe that the loader initiates:

   • Downloading audio files.
   • Extracting and building audio clips.

---

Expected Behavior:

When input_mode=InputMode.TEXT_ONLY is set, the loader should:

• Skip downloading audio (download_audio)
• Skip generating clips (generate_clips)
• Avoid referencing audio paths like snippet_paths and dialogue_paths

---

Actual Behavior:

Despite requesting text-only mode:

• The loader downloads audio via self.download_audio() in load().
• Audio clips are generated via self.generate_clips().

These steps are triggered unconditionally if audio_clips does not already exist

if not self.clips_path.exists():
    logging.info('Downloading audio data...')
    self.download_audio()
    logging.info('Download completed!')

    logging.info('Building audio clips...')
    self.generate_clips()
    logging.info('Build completed')

    shutil.rmtree(self.audio_path)

---

Suggested Fix:

Conditionally skip the audio download/clip generation in load() based on the input_mode value:

if self.input_mode != InputMode.TEXT_ONLY and not self.clips_path.exists():
    ...

---