Init shortcut job

[lhoestq]

Add the first jobs that allows shortcuts, this will be useful to accelerate the viewer's preparation for Parquet datasets (run all jobs at once)

Details

• Job runners can now yield multiple job results, including ShortcutJobResult objects that shortcut other runners
• The AfterJobPlan takes this into account and creates new jobs based on the job result and the shortcut job results
• there is a new "dataset-init" job runner that shortcuts "dataset-config-names" to start with
• in a second step (future PR), "dataset-init" will also be able to shortcut Parquet-related job runners
• later it can also be useful to group jobs with low difficutly (e.g. info aggregation jobs when there are many configs and splits) and shortcut them as well, greatly reducing the number of jobs in the queue

Thanks to open models for the help coding this :p

struggling a bit to merge that one due to tests failing bc of a mix of network/github/hub-ci/app-tokens issues

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.