Handle differnt dtype in MultiNestedTensor/MultiEmbeddingTensor

[akihironitta]

num_rows = 10
tensor_list = [
    torch.rand((num_rows, 2, dtype=torch.float32),
    torch.rand((num_rows, 2, dtype=torch.long),
]
MultiEmbeddingTensor.from_list(tensor_list)  # should raise an exception or warning

---

You probably want to check if all items in xs are in the same dtype. If not, probably good to raise an error or a warning.

Originally posted by @yiweny in #181 (comment)

[akihironitta]

We should either error out, raise a warning or simply documenting the behaviour, linking how PyTorch handles.

[weihua916]

Good point! I think we can raise an informative error.

[akihironitta]

I feel like just documenting the behaviour is sufficient because PyTorch already makes it clear about type promotion https://pytorch.org/docs/stable/tensor_attributes.html#torch-dtype wdyt? (cc @yiweny)