Fixed bug in the action scale of the intervention actions and offline dataset actions. (scale by inverse delta)

Co-authored-by: Adil Zouitine <adizouitinegm@gmail.com>

lerobot/scripts/server/learner_server.py

@@ -600,6 +600,7 @@ def train(cfg: DictConfig, out_dir: str | None = None, job_name: str | None = No
             device=device,
             state_keys=cfg.policy.input_shapes.keys(),
             action_mask=active_action_dims,
+            action_delta=cfg.env.wrapper.delta_action,
         )
         batch_size: int = batch_size // 2  # We will sample from both replay buffer