Support different audio / directory structures

### Feature Description

Support a more direct directory structure of speakers instead of "conversations". i.e.:

```
data/
├── bob/
| ├── 0.wav
| ├── 1.wav
| ├── 2.wav
| ├── 3.wav
| ├── 4.wav
├── sally/
| ├── 5.wav
| ├── 6.wav
| ├── 7.wav
| ├── 8.wav
| ├── 9.wav
└── eva/
├── 10.wav
├── 11.wav
├── 12.wav
├── 13.wav
├── 14.wav
```

Where all the audio for each speaker is provided as a directory. This would involve creating new functions for preparing the dataset -- with no guarantee that the "conversation id" holdout condition is met.

### Use Case

See #17 -- direct use case already done.

### Solution

### Alternatives

Start by reviewing issue #17 and the existing dataset-preparation flow in the repository. Determine how the current conversation-based structure and conversation-id holdout are handled, then define the behavior for speaker directories without that holdout guarantee. Done means the documented speaker-directory layout is accepted and the resulting dataset preparation behavior is covered.

Written by the indexing model from the issue text.