Reduce range size shape outputs on abstract/concrete mixtures

### Example

([Playground link](https://carrascomj.github.io/shapelsbook/playground#eNrLzC3ILypRKMkvSs7g4spWsFUwNOCqAFJgEb3EosS89FSNbE0uABZ4DLo))

```python
import torch

idx = torch.arange(k)
```

Hovering `idx` yields a shape `[k-0/1]`, which can be reduced to `[k]`, as done already for (trans)convolutions and pooling (and it's missing parentheses, anyways.)

Start by tracing how the shape for torch.arange is produced for abstract/concrete mixtures, then compare it with the existing trans convolution and pooling reductions. Done means the hover output for idx shows [k] instead of [k-0/1] and does not include the noted missing parentheses.

Written by the indexing model from the issue text.