- Refactor observation encoder in modeling_sac.py

- added `torch.compile` to the actor and learner servers.
- organized imports in `train_sac.py`
- optimized the parameters push by not sending the frozen pre-trained encoder.

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

lerobot/configs/policy/sac_maniskill.yaml

@@ -52,6 +52,8 @@ policy:
   n_action_steps: 1
 
   shared_encoder: true
+  # vision_encoder_name: null
+  freeze_vision_encoder: false
   input_shapes:
     # # TODO(rcadene, alexander-soare): add variables for height and width from the dataset/env?
     observation.state: ["${env.state_dim}"]