Make the behavior of changing election_timeout_ms clearer

[andrwng]

Who is this for and what problem do they have today?

When changing election_timeout_ms there's no indication to the user that the cluster needs to be restarted. This is technically true, but in practice, this property gets changed when an existing cluster is overloaded, and there's an expectation from admins that it just works. We should make it less surprising when changing this doesn't affect the number of leader step downs.

What are the success criteria?

Either

• Make the property restart_required::true
• Reset the jitters when the property is changed (though there may be some gotchas here, e.g. if replicas on different nodes see different properties)

Why is solving this problem impactful?

Improved operational experience

JIRA Link: CORE-1550

[daisukebe]

I believe this was addressed in #19899