If I have 3 masters and accidentally set minimum_master_nodes to 4 dynamically, the cluster will stop. While the cluster is stopped, there's no way to update settings, so I have to do a full restart.

If I updated a persistent setting, then not even a full restart will fix the issue. I have to manually edit cluster state.

You could get out of this situation by adding master nodes until the new minimum is reached, but that's not a complete solution because I might have fat fingered minimum_master_nodes to 100 or something like that.

I think it'd be worth adding some validation to ensure that an update to minimum_master_nodes won't accidentally put the cluster in an unrecoverable state.