Allow load_dataset to take a working dir for intermediate data

### Feature request

As a user, I can set a working dir for intermediate data creation. The processed files will be moved to the cache dir, like
```
load_dataset(…, working_dir=”/temp/dir”, cache_dir=”/cloud_dir”).
```

### Motivation

This will help the use case for using datasets with cloud storage as cache. It will help boost the performance.

### Your contribution

I can provide a PR to fix this if the proposal seems reasonable.