[core] Don't log actor restart warning if arg is detached actor

[dayshah]

Description

Ray Data doesn't want this warning log to show on all workloads. Now, they pass a detached ActorLocationTracker into actors with max_restarts > 0. A detached actor won't go out of scope, so as long as it isn't ray.killed it'll still exist when the actor with it as an arg needs to restart. See - #57838 - minimal ray data usage will result in the log now.

We'll revert this logic back once #53727 is fixed.

Here I'm just checking if an inlined arg passed by ref is a detached actor. If it is, we'll skip setting the warning flag so we won't log the warning. The actor handle has no way of knowing whether the actor is detached so I had to create that path.

Used this script to test that we don't log the warning.

@ray.remote(max_restarts=-1)
class Actor:
    def __init__(self, other_actor_ref):
        self.other_actor_ref = other_actor_ref

    def do_thing(self):
        print("hi")

detached_actor = Actor.options(name="detached_actor").remote(None)
actor = Actor.remote(detached_actor)

ray.get(actor.do_thing.remote())

Note: There's one gotcha here - is_detached doesn't get through all the way if you're passing the actor handle around because the serialization doesn't take it into account, I'll fix this in a follow up right after