Fix init_index to create shards in audio/ subdirectory

[Copilot]

list_all_shards() only discovers shards within subdirectories, not at the dataset root. When init_index=True, shards written directly to dataset_root resulted in an empty index.

Changes

• Redirect shard output when init_index=True: If output_dir is not already named "audio", automatically write shards to output_dir/audio/ subdirectory
• Preserve existing behavior: When output_dir already ends with "audio", write shards directly there (avoids nested audio/audio/)
• Set dataset_root correctly: Compute before potential redirection to ensure key_mapping.json and index.sqlite3 are placed at the proper root level

Example

# Before: shards written to dataset_root, empty index created
shard_from_audio_dir("input/", "my_dataset/", init_index=True)
# my_dataset/audio-00000.wsds  ← not discoverable
# my_dataset/index.sqlite3      ← empty

# After: shards written to audio/ subdirectory, index populated correctly  
shard_from_audio_dir("input/", "my_dataset/", init_index=True)
# my_dataset/audio/audio-00000.wsds  ← discoverable
# my_dataset/index.sqlite3            ← populated

# Already-named audio/ dir works as before
shard_from_audio_dir("input/", "my_dataset/audio/", init_index=True)
# my_dataset/audio/audio-00000.wsds  ← no nested audio/audio/
# my_dataset/index.sqlite3

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.