POC: add runtime teleoperation policy

[MarkRedeman]

Reopening #131 on main branch.

TL/DR: implement a model policy whose predicted actions are the observed state from a leader robot.

class TeleoperatorPolicy:
    def predict_action_chunk(self, observation: dict[str, np.ndarray]) -> np.ndarray:  # noqa: ARG002
        """Read the leader and return one action for the follower."""
        leader_obs = self.leader_robot.get_observation()
        action = np.asarray(leader_obs.joint_positions, dtype=np.float32)
        return action[None, :]

To do this we change the InferenceModel in our runtime and execution classes to be a RuntimePolicy, which is a protocol that's implemented by both InferenceModel and TeleoperatorPolicy.

I'm reopening this mainly to show how teleoperation could be done, though it might be better to spend a bit more time on the interface design for this. Another open that we would need to look into is see if we can add a RecordingCallback so that we can record a teleoperation session.