Missing Tensor features

[EricLBuehler]

Hello everybody,

I have identified several missing features from Tensor during my development of candle-vllm and candle-sampling. This issue serves to document these missing features:

• Slice fill is missing
  It appears that there is no way to fill a slice of a tensor with values after creation. Possible solutions could include from a Copy type or a closure.
• Iteration
  Tensor does not provide a method to iterate over a slice in the tensor.

I will update this issue if I find more apparently missing features.

Please correct me if these features are actually implemented. @LaurentMazare, do you think that these features could be added? It appears that Pytorch has these features implemented, so I think it is possible.