Represent slices natively in traces

Nobody has claimed this yet.

## 🚀 Feature

### Motivation

Tensor slices are represented in traces as:
```
t107 = torch_slice_prim_impl(t53, [0, 0, 0, 0], [4, 32, 2048, 0], [1, 1, 1, 1]) # t107: "cuda:0 bf16[4, 32, 2048, 0]"
```

But there's no `torch_slice_prim_impl` import. And we can use Python to represent it.

This reference comes from:

https://github.com/Lightning-AI/lightning-thunder/blob/ea1d1302f4a630e3832c07dc3adfe559111ba099/thunder/executors/torchex.py#L533-L534

https://github.com/Lightning-AI/lightning-thunder/blob/ea1d1302f4a630e3832c07dc3adfe559111ba099/thunder/executors/torchex.py#L507-L517

### Pitch

Instead represent it with `__getitem__` and `slice()`:

```python
t123 = t321.__getitem__([slice(0, 3), slice(0, 5)]) # t123: "cuda:..."
```

### Alternatives

Add the `torch_slice_prim_impl` import from torchex to the trace so that it's a valid program

cc @apaz-cli @nikitaved

Start by reading the referenced sections of thunder/executors/torchex.py, especially the slice implementation and related trace handling. Compare the current torch_slice_prim_impl representation with the proposed __getitem__ and slice() form. Done means slices are emitted as valid native trace expressions, or the alternative import is explicitly implemented.

Written by the indexing model from the issue text.