Multi GPU support

Hi @1adrianb, thank you for publishing such an amazing library. I was wondering if it would be possible to add support for running on multiple GPUs simultaneously?

The issue requests multi-GPU support for the face-alignment library. Look at the main inference or training scripts to understand the current single-GPU setup. Research PyTorch's DataParallel or DistributedDataParallel modules for implementing multi-GPU. Check if there are existing tests for GPU usage to validate the changes.

索引モデルが issue の本文から書いたものです。