Fixed AuthorityRound deadlock on shutdown

[debris]

closes #8088

follow up for @svyatonik explanation

memory in our event loop looks as follows:

IoService -> manager_thread -> IoManager -> Workers -> workers_threads

and in case of AuthorityRound it was additionally:

workers_threads -> TransitionHandler -> Weak<AuthorityRound> -> IoService

everything was fine, as long as we were not trying to shutdown one of the workers when he owned the strong reference of IoService

as a solution, I refactored the code so TransitionHandler thread never has ownership of IoService

[svyatonik]

Could confirm that the issue is fixed. Code changes also LGTM (no more engine.upgrade() calls from TransitionHandler).

[ordian]

Can not reproduce #8088 on Kovan with this changes.

[ordian]

Maybe it's worth extracting into a function:

fn step(step: &PermissionedStep, client: &RwLock<Option<Weak<EngineClient>>>) {
	step.inner.increment();
	step.can_propose.store(true, AtomicOrdering::SeqCst);
	if let Some(ref weak) = *client.read() {
		if let Some(c) = weak.upgrade() {
			c.update_sealing();
		}
	}
}

to avoid code duplication?

[debris]

this code would also leak, cause it upgrades the reference to EngineClient from worker thread

[rphmeier]

Engine's step function might be completely unnecessary now. it's currently used for tests but we can use TransitionHandler in the same way. Not a blocker for this PR but we could file to refactor.