Should `ManagerAPI.cancel()` raise an error if the service is already stopping or finished

from: https://github.com/ethereum/async-service/pull/31#discussion_r359818352

What is wrong

Currently, #31 introduces the rule that if a service is already stopping or finished that calling ManagerAPI.cancel() will return early without doing anything.

Should this instead raise a LifecycleError?

The counter argument to this would be a case where we want to be able to blindly call manager.cancel(). If we choose to raise an exception then the call site would have to do this conditionally which adds boilerplate.

The argument for this is that removes the ambiguity where ManagerAPI.is_cancelled will not always be True after a call to Manager.cancel().

How can it be fixed

Decide which of the above is more important.

Review PR #31 and its discussion, then inspect the ManagerAPI.cancel(), ManagerAPI.is_cancelled, and LifecycleError entry points. Decide whether cancellation after stopping or finishing should be a no-op or raise; done means the chosen behavior is explicit and consistently reflected in the lifecycle handling.