Ability to update a dataset

### Feature request

The ability to load a dataset, add or change something, and save it back to disk.

Maybe it's possible, but I can't work out how to do it, e.g. this fails:

```py
import datasets

dataset = datasets.load_from_disk("data/test1")
dataset = dataset.add_item({"text": "A new item"})
dataset.save_to_disk("data/test1")
```

With the error:
```
PermissionError: Tried to overwrite /mnt/c/Users/david/py/learning/mini_projects/data_sorting_and_filtering/data/test1 but a dataset can't overwrite itself.
```

### Motivation

My use case is that I want to process a dataset in a particular way but it doesn't fit in memory if I do it in one go. So I want to perform a loop and at each step in the loop, process one shard and append it to an ever-growing dataset. The code in the loop will load a dataset, add some rows, then save it again.

Maybe I'm just thinking about things incorrectly and there's a better approach. FWIW I can't use `dataset.map()` to do the task because that doesn't work with `num_proc` when adding rows, so is confined to a single process which is too slow.

The only other way I can think of is to create a new file each time, but surely that's not how people do this sort of thing.

### Your contribution

na