Describe the workflow you want to enable

The purpose of this request is to limit the number of random failures we get in automated executions (e.g. on our CI) caused by unreliable network connections.

Describe your proposed solution

Add a n_retries=3 parameter to fetch_openml. In case of HTTPError, fetch_openml would automatically retry up to n_retries times, waiting 1s between each trial.

Other fetch_* methods could benefit from a similar option.