Provide a way to write a streamed dataset to the disk

**Is your feature request related to a problem? Please describe.**

The streaming mode allows to get the 100 first rows of a dataset very quickly. But it does not cache the answer, so a posterior call to get the same 100 rows will send a request to the server again and again.

**Describe the solution you'd like**

Provide a way to write the streamed rows of a dataset on the disk, and to load from it later.

**Describe alternatives you've considered**

Provide a third mode: `lazy`, which would use the local cache for the data that have already been fetched previously, and use streaming to get the rest of the requested data.