[Feature] Include Support for Trainium and Inferentia

### Feature request As of today, Oumi doesn't seem to support AI specific chips from AWS, namely Traininum and Inferentia. They don't show as available GPUs, and the process of training is given to the CPU in these cases. ### Motivation / references Enabling support for these can make Oumi more open for different cloud training GPU options. Relevant documentation for this is [available here](https://awsdocs-neuron.readthedocs-hosted.com/en/latest/) probably [here as well](https://github.com/huggingface/optimum-neuron/tree/main), and also at [huggingface](https://huggingface.co/docs/optimum-neuron/main/guides/fine_tune). ### Your contribution I can try to help in this matter, but I'm not a huge expert in this area, and it would involve some expert in the training code from `transformers` which seems to be the main train path (at least for SFT).