Additional Tensor Operations?

I am trying to perform the following tensor operations [torch.cumsum()](https://pytorch.org/docs/stable/generated/torch.cumsum.html), and the quickest way i can think of is to convert the tensor to a vec and perform the operation using

```rust
fn cumsum_2d(mask: &Tensor, dim: u8, device: &Device) -> Result {
let mask = mask.to_vec2::()?;

let rows = mask.len();
let cols = mask[0].len();

let mut result = mask.clone();

match dim {
0 => {
// Cumulative sum along rows
for i in 0..rows {
for j in 1..cols {
result[i][j] += result[i][j - 1];
}
}
}
1 => {
// Cumulative sum along columns
for j in 0..cols {
for i in 1..rows {
result[i][j] += result[i - 1][j];
}
}
}
_ => panic!("Dimension not supported"),
}

let result = Tensor::new(result, &device)?;

Ok(result)
}
```

@LaurentMazare could you recommend a more effective way to perform this operation? and maybe `torch.split(tensor, split_size_or_sections, dim=0)` too..
Thanks!

Start by reviewing the Tensor API around to_vec2 and Tensor::new, then compare the linked torch.cumsum documentation. Clarify whether cumsum and torch.split are both in scope; done should mean these operations work without converting tensors to Vec values and their behavior is covered by tests.

Written by the indexing model from the issue text.