Allow controlling scheduler jobs from disabled cluster nodes

Since XWIKI-22855, the recommendation in the case of a clustering is to disable the Scheduler on all instances but one, and you then cannot do much in the disabled instances.

It would be interesting to refactor a bit how the scheduler in working to allow all nodes to control jobs, even if the scheduler is disabled for them. It mainly means relying on serializable events instead of direct calls (that way, the main instance can pick those events and interpret them wherever they are coming from).

Now, this might also be the occasion to introduce a whole new module/API to replace the old plugin which we'll have to get rid of eventually (like any plugin).