Add human intervention mechanism and eval_robot script to evaluate policy on the robot

[michel-aractingi]

What this does

Related to the human intervention category in issue #504 . What this does:

• Adds an eval_robot.py script to run rollouts of the policy on the real robot.
• Adds the ability to use the keyboard to pause the policy actions.
• Adds the ability to allow the user to take over the policy by using the leader arm to control the follower through teleoperation.

How it was tested:

In the main of the eval_robot.py we run the eval_policy function that runs the robot's action in a fixed command (in the future this will be the policy actions). Once the space is pressed the actions are paused and message is written for the human to prepare the leader in a similar pose to take over. Once another space is pressed the teleoperation begins. The user can press the right arrow to move on to the next rollout or the left arrow to rerecord the episode.

you can test it yourself by:

python lerobot/scripts/eval_robot.py --robot-path lerobot/configs/robot/koch.yaml

TODOs

• fix the timing when the policy is paused so that the time is not counted with the control time.
• add reward classifier from Reward classifier and training #528 to label the rewards of the rollouts.
• add mechanism to classify success
• test exploration with policy

[michel-aractingi]

cc @ChorntonYoel