Add multilingual dataset eval support

[arda-argmax]

After installing whisperkittools, you can run the following command to run evaluation on common voice 17 dataset:

whisperkit-evaluate-model --model-version <model_version> --output-dir <output_dir> --evaluation-dataset common_voice_17_0-argmax_subset-400 --pipeline WhisperKit --num-proc 1

Example CLI command

whisperkit-evaluate-model --model-version openai/whisper-tiny --output-dir out/ --evaluation-dataset common_voice_17_0-argmax_subset-400 --pipeline WhisperKit --num-proc 1

Additional args:

  --force-language      If specified, forces the language in each data sample
                        (if available)
  --language-subset LANGUAGE_SUBSET
                        If specified, filters the dataset for the given
                        language

To use only the en subset of the dataset, add --language-subset en CLI arg
To force the transcription task and language, add --language-subset <lang> --force-language