Similar issue as solved in #78829.

This API is surprisingly expensive because it requests the full metadata via get cluster state request from master when called only to return just the settings from it to the user. This can mean requesting tens or hundreds of MB from master in a large cluster unexpectedly.

We should adjust this to not require fetching the full cluster state. One option would be to add a dedicate master node action for returning just the settings.

Relates #77466