Add support for xla devices

[BitPhinix]

What does this PR do?

Adds support for xla devices via torch_xla (https://pytorch.org/xla/release/2.3/index.html)

Test code:

import torch
import torch_xla.core.xla_model as xm
from safetensors.torch import save_file
from safetensors import safe_open

test = {"a": torch.tensor([1, 2, 3])}

save_file(test, 'test.safetensors')

with safe_open("test.safetensors", framework="pt", device="xla") as f:
   for key in f.keys():
      print(f.get_tensor(key).device)

[Narsil]

torch_xla had a lot of issues and different API based on the actual accelerator.

Given the narrow PR scope however I feel like this is a good addition if that enables XLA.
But this will not be tested.