[core] Add __ray_call__ default actor method

[ericl]

Why are these changes needed?

Similar to __ray_ready__, this adds a developer method __ray_call__(fn, *a, **kw) that can be used by library developers to run custom closures on actors.

For example, a common use case for an actor is to "get its current node id". Without this method, a library developer would have to tell users to add a special get_node_id method to their actor. With this new API, the library developer can instead run:

def fn(actor_self):
   return ray.get_runtime_context().get_node_id()

ray.get(actor.__ray_call__.remote(fn))